def getConnection(is_prod = False): if is_prod: mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) else: mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=SANDBOX_HOST) print mtc.get_account_balance() return mtc
class MTurk(object): def __init__(self): self.mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) def balance(self): return self.mtc.get_account_balance() def create_question(self, id, title): content = QuestionContent() content.append_field('Title', title) text = FreeTextAnswer() return Question(identifier=id, content=content, answer_spec=AnswerSpecification(text)) def create_hit(self, title, description=''): question_form = QuestionForm() question_form.append(self.create_question(id='1', title='Comments')) question_form.append(self.create_question(id='2', title='More Comments')) self.mtc.create_hit(questions=question_form, max_assignments=1, title=title, description=description, duration=60*5, reward=0.01) def external(self): q = ExternalQuestion(external_url="http://mturk-hit-wizard.herokuapp.com/view/2", frame_height=800) #conn = MTurkConnection(host=HOST) keywords=['boto', 'test', 'doctest'] create_hit_rs = self.mtc.create_hit(question=q, lifetime=60*65,max_assignments=2,title="Boto External Question Test", keywords=keywords,reward = 0.05, duration=60*6,approval_delay=60*60, annotation='An annotation from boto external question test', response_groups=['Minimal','HITDetail','HITQuestion','HITAssignmentSummary',]) assert(create_hit_rs.status == True)
def con(HOST=HOST): try: mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) print('Connected to '+HOST) print('Balance: '+str(mtc.get_account_balance())) return mtc except: print('Failed to connect to '+HOST+', plz make sure you are connecting to the correct host with correct credentials')
def test(): text = ("Two young men with backpacks walked with purpose down " "Boylston Street Monday afternoon, weaving through the " "crowd on the sidelines of the Boston Marathon. It seemed " "like they'd been there before, like they knew where they " "were going.") title = "From backpacks to 'flash-bangs': Boston's week of terror" url = "http://www.cnn.com/2013/04/21/us/boston-week-review/?hpt=hp_t1" fact = MTurkSurveyFactory() questionForms = fact.buildSurvey([[url, title, text]]) print 'getting account balance' mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST, is_secure=True, https_connection_factory=(https_connection_factory, ())) fact.submitHITs(mtc=mtc, questionForms=questionForms) print mtc.get_account_balance()
def test(): text = ("Two young men with backpacks walked with purpose down " "Boylston Street Monday afternoon, weaving through the " "crowd on the sidelines of the Boston Marathon. It seemed " "like they'd been there before, like they knew where they " "were going.") title = "From backpacks to 'flash-bangs': Boston's week of terror" url = "http://www.cnn.com/2013/04/21/us/boston-week-review/?hpt=hp_t1" fact = MTurkSurveyFactory() questionForms = fact.buildSurvey([[url, title, text]]) print 'getting account balance' mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST, is_secure=True, https_connection_factory=(https_connection_factory, ())) fact.submitHITs(mtc=mtc, questionForms=questionForms) print mtc.get_account_balance()
def connect_AMT(): # Open connections to HOST # using ACCESS_ID,SECRET_KEY # obtain from https://portal.aws.amazon.com/gp/aws/securityCredentials#access_credentials mturk_conn = None try: mturk_conn = MTurkConnection(aws_access_key_id=ACCESS_ID,aws_secret_access_key=SECRET_KEY,host=HOST) #will throw if not connected canary = mturk_conn.get_account_balance() print "Connection to amt established." except Exception as e1: print "[Error Connecting]",e1 print "[Exiting]" exit(1) return mturk_conn
def connect_AMT(): # Open connections to HOST # using ACCESS_ID,SECRET_KEY # obtain from https://portal.aws.amazon.com/gp/aws/securityCredentials#access_credentials mturk_conn = None try: mturk_conn = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) # will throw if not connected canary = mturk_conn.get_account_balance() print "Connection to amt established." except Exception as e1: print "[Error Connecting]", e1 print "[Exiting]" exit(1) return mturk_conn
class TurkConnect: def __init__(self, host, n=10): mturkparams = dict( aws_access_key_id=config.get('AWS Access', 'aws_access_key_id'), aws_secret_access_key=config.get('AWS Access', 'aws_secret_access_key'), host=host) self.mtc = MTurkConnection(**mturkparams) # Configure portal experimentPortalURL = config.get('HIT Configuration', 'question_url') frameheight = 600 mturkQuestion = ExternalQuestion(experimentPortalURL, frameheight) # Qualification: quals = Qualifications() approve_requirement = config.getint('HIT Configuration', 'Approve_Requirement') quals.add( PercentAssignmentsApprovedRequirement("GreaterThanOrEqualTo", approve_requirement)) if config.getboolean('HIT Configuration', 'US_only'): quals.add(LocaleRequirement("EqualTo", "US")) # Specify all the HIT parameters self.paramdict = dict( hit_type=None, question=mturkQuestion, lifetime=datetime.timedelta( hours=config.getfloat('HIT Configuration', 'HIT_lifetime')), max_assignments=config.getint('HIT Configuration', 'max_assignments'), title=config.get('HIT Configuration', 'title'), description=config.get('HIT Configuration', 'description'), keywords=config.get('HIT Configuration', 'keywords'), reward=config.getfloat('HIT Configuration', 'reward'), duration=datetime.timedelta( hours=config.getfloat('HIT Configuration', 'duration')), approval_delay=None, questions=None, qualifications=quals) def checkbalance(self): return self.mtc.get_account_balance() # Tests the connection def createhit(self): myhit = self.mtc.create_hit(**self.paramdict)[0] hitid = myhit.HITId
class TurkConnect: def __init__(self, host, n=10): mturkparams = dict( aws_access_key_id = config.get( 'AWS Access', 'aws_access_key_id' ), aws_secret_access_key = config.get( 'AWS Access', 'aws_secret_access_key' ), host=host) self.mtc = MTurkConnection( **mturkparams ) # Configure portal experimentPortalURL = config.get( 'HIT Configuration', 'question_url' ) frameheight = 600 mturkQuestion = ExternalQuestion( experimentPortalURL, frameheight ) # Qualification: quals = Qualifications(); approve_requirement = config.getint('HIT Configuration', 'Approve_Requirement') quals.add( PercentAssignmentsApprovedRequirement("GreaterThanOrEqualTo", approve_requirement)) if config.getboolean('HIT Configuration', 'US_only'): quals.add( LocaleRequirement("EqualTo", "US") ) # Specify all the HIT parameters self.paramdict = dict( hit_type = None, question = mturkQuestion, lifetime = datetime.timedelta(hours=config.getfloat('HIT Configuration', 'HIT_lifetime')), max_assignments = config.getint('HIT Configuration', 'max_assignments'), title = config.get('HIT Configuration', 'title'), description = config.get('HIT Configuration', 'description'), keywords = config.get('HIT Configuration', 'keywords'), reward = config.getfloat('HIT Configuration', 'reward'), duration = datetime.timedelta( hours=config.getfloat('HIT Configuration', 'duration')), approval_delay = None, questions = None, qualifications = quals ) def checkbalance(self): return self.mtc.get_account_balance() # Tests the connection def createhit(self): myhit = self.mtc.create_hit( **self.paramdict )[0] hitid = myhit.HITId
def check_account_balance(num_instances, price): """ Checks if the requester's mturk account has enough balance to post hits. Args: num_instances (int) : Number of instances that will be posted on Mturk price (float) : Reward for each correct answer Returns: 1 : If account has enough balance -1 : If account has insuffcient balance """ mtc = MTurkConnection(aws_access_key_id = aws_parameters['access_key'], aws_secret_access_key = aws_parameters['secret_key'], debug = 1, host = DEV_HOST) #print mtc.get_account_balance() if mtc.get_account_balance() < num_instances * price: return -1 else: return 1
def new_rate_hit(PIN_IMAGE_URL, PIN_IMAGE_TITLE, MACYS_IMAGE_URL, MACYS_IMAGE_TITLE): mtc = MTurkConnection(aws_access_key_id=AWS_ACCESS_KEY_ID, aws_secret_access_key=AWS_SECRET_ACCESS_KEY, host=HOST) if debug: print mtc.get_account_balance() title = 'Match these Pictures to Macy\'s Products' description = 'Look at this photo and match it to Macy\'s products' keywords = 'clothing, rating, opinions, easy, quick, macys' ratings = [('Very Bad', '1'), ('Bad', '2'), ('OK', '3'), ('Good', '4'), ('Very Good', '5')] #make overview overview = Overview() overview.append_field('Title', 'Rank how these two images match.') overview.append( FormattedContent('<table border="1">><tr><td width="50%"><img src="' + PIN_IMAGE_URL + '" alt="Pintrest Image" /></td>' '<td width="50%"><img src="' + MACYS_IMAGE_URL + '" alt="Macys Image" /></td></tr><tr>' '<td width="50%">' + PIN_IMAGE_TITLE + '</td><td width="50%">' + MACYS_IMAGE_TITLE + '</td></tr></table>')) #make q1 qc1 = QuestionContent() qc1.append_field('Title', 'Rank the match between these two') fta1 = SelectionAnswer(min=1, max=1, style='dropdown', selections=ratings, type='text', other=False) q1 = Question(identifier='rating', content=qc1, answer_spec=AnswerSpecification(fta1), is_required=True) #make q2 qc2 = QuestionContent() qc2.append_field('Title', 'Comments about the HIT (Optional)') fta2 = FreeTextAnswer() q2 = Question(identifier="comments", content=qc2, answer_spec=AnswerSpecification(fta2)) #make question form question_form = QuestionForm() question_form.append(overview) question_form.append(q1) question_form.append(q2) #--------------- CREATE THE HIT ------------------- mtc.create_hit(questions=question_form, max_assignments=1, title=title, description=description, keywords=keywords, duration=60 * 5, reward=0.05)
created_hits = [] for q in questions: try: hit = mtc.create_hit(question=q, max_assignments=hitdata['assignments'], title=hitdata['title'], description=hitdata['description'], keywords=hitdata['keywords'], duration=duration, lifetime=lifetime, approval_delay=approvaldelay, reward=reward, qualifications=quals) created_hits.append(hit) except MTurkRequestError as e: print('{}: {}\n{}'.format(e.status, e.reason, e.body)) hit_list = [{'HITId': y.HITId, 'HITTypeId': y.HITTypeId} for y in [x[0] for x in created_hits]] outfilename = hitfile_name.split('.') outfilename.insert(-1, 'success') outfilename = '.'.join(outfilename) with open(outfilename, 'w') as successfile: safe_dump(hit_list, stream=successfile, default_flow_style=False) preview_url = 'https://workersandbox.mturk.com/mturk/preview?groupId={}' if args.sandbox else 'https://www.mturk.com/mturk/preview?groupId={}' for hittypeid in {x['HITTypeId'] for x in hit_list}: print('You can preview your new HIT at:\n\t{}'.format(preview_url.format(hittypeid))) print('{0} is the final balance'.format(mtc.get_account_balance()))
class MTurkClient: # SETUP # =========== def __init__(self,aws_access_key,aws_secret_key,aws_mode): self.mode = aws_mode if aws_mode == 'sandbox': self.host = 'mechanicalturk.sandbox.amazonaws.com' else: self.host = 'mechanicalturk.amazonaws.com' self.c = MTurkConnection( aws_access_key, aws_secret_key, host=self.host) default_settings = { 'lifetime': DAY, 'duration': 10 * MINUTE, 'approval_delay': DAY, 'title': "[title]", 'description': "[description]", 'keywords': [], 'reward': 0.01, 'max_assignments': 1, 'height': 700, 'qualifications': [], } # HITS # =========== def create_hit(self,url,extra_settings): "Eventually, this should take a TEMPLATE and a dictionary of INPUT data that's put into that template. This function would then create an HTML file locally (assuming we're running on a web server) by replacing template {tags} with input values, and then send the URL to the newly created page to MTurk." settings = self.default_settings.copy() settings.update(extra_settings) settings['reward'] = Price(settings['reward']) settings['qualifications'] = qualification.Qualifications(settings['qualifications']) settings['keywords'] = ','.join(settings['keywords']) height = settings.pop('height') hit = self.c.create_hit(question=ExternalQuestion(url,height),**settings)[0] #print 'Created hit %s' % hit.HITId return hit.HITId,hit.HITTypeId #hit_type=None, # Let Amazon do this automatically #annotation=None, # Optional annotation for our system to use #questions=None, # If you want to create multiple HITs at a time? Probably irrelevant for External #response_groups=None, # Unclear what this does def get_hit(self,hit_id): return self.c.get_hit(hit_id)[0] def hit_results(self,hit_id,type=None): # type in ['Submitted','Approved','Rejected',None] results = {} assignments = self.c.get_assignments(hit_id, status=None, page_size=100) for asst in assignments: results.setdefault(asst.AssignmentId,{}) answers = asst.answers[0] for qfa in answers: field, response = qfa.qid, qfa.fields[0] results[asst.AssignmentId][field] = response results[asst.AssignmentId]['worker_id'] = asst.WorkerId results[asst.AssignmentId]['accept_time'] = datetime.strptime(asst.AcceptTime,"%Y-%m-%dT%H:%M:%SZ") results[asst.AssignmentId]['submit_time'] = datetime.strptime(asst.SubmitTime,"%Y-%m-%dT%H:%M:%SZ") return results # URL of a HIT on MTurk def hit_url_turk(self,hit_id): pass def hit_url_external(self,hit_id): pass def extend_hit(self,hit_id,extras): return self.c.extend_hit(hit_id, extras) @catcherror def delete_hit(self,hit_id): self.c.disable_hit(hit_id) # Deletes all the HITS on the server. Risky! def cleanup(self): for hit in self.c.get_all_hits(): self.delete_hit(hit.HITId) # ASSIGNMENTS # =========== @catcherror def approve(self, asst_id, feedback=None): return self.c.approve_assignment(asst_id, feedback) @catcherror def reject(self, asst_id, feedback=None): return self.c.reject_assignment(asst_id, feedback) def block(self,worker_id,feedback=None): return self.c.block_worker(worker_id, feedback) def unblock(self,worker_id,feedback=None): return self.c.unblock_worker(worker_id, feedback) def bonus(self,asst,amount,feedback): return self.c.grant_bonus(asst.worker, asst.asst_id, Price(amount), feedback) # STATUS / DIAGNOSTICS # -------------------- def balance(self): return self.c.get_account_balance()[0]
import os print os.getcwd() import sys import boto from boto.mturk.connection import MTurkConnection SAND = 0 ACCESS_ID = 'AKIAIM5D5I7RUTGYNI7A' SECRET_KEY = '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' HIT = "2JNL8I9NZW6HG96GKYHWCT87ATCVL9" # https://mechanicalturk.amazonaws.com/?Service=AWSMechanicalTurkRequester HOST = 'mechanicalturk.amazonaws.com' mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) print mtc.get_account_balance() mtc.extend_hit(hit_id=HIT, assignments_increment=1)
class MTurk(object): """ A class that wraps a boto.mturk.connection object and provides methods for the most common AI2 use cases """ def __init__(self, aws_access_key_id, aws_secret_access_key, host=SANDBOX_HOST): """ initializes the instance with AWS credentials and a host :param aws_access_key_id the access key id. :param aws_secret_access_key the secret access key. :param host the mturk host to connect to """ self.connection = MTurkConnection( aws_access_key_id=aws_access_key_id, aws_secret_access_key=aws_secret_access_key, host=host) self.host = host def __del__(self): """ close the connection whenever this object goes out of scope """ self.connection.close() def get_account_balance(self): """ :return the balance on the mturk account """ return self.connection.get_account_balance()[0] def _create_hit(self, params, **kwargs): """ internal helper function for creating a HIT :param params the parameters (required and optional) common to all HITs :param **kwargs any other parameters needed for a specific HIT type :return the created HIT object """ return self.connection.create_hit( title=params["title"], description=params["description"], keywords=params["keywords"], max_assignments=params["max_assignments"], reward=Price(amount=params["amount"]), qualifications=params["qualifications"], lifetime=params["lifetime"], # optional params below annotation=params.get("annotation"), **kwargs) def create_url_hit(self, params): """ creates a HIT for an external question with a specified URL :param params a dict of the HIT parameters. must contain a "url" parameter :return the created HIT object """ question = ExternalQuestion(params["url"], params["frame_height"]) return self._create_hit(params, question=question) def create_html_hit(self, params): """ creates a HIT for a question with the specified HTML :param params a dict of the HIT parameters, must contain a "html" parameter :return the created HIT object """ question = HTMLQuestion(params["html"], params["frame_height"]) return self._create_hit(params, question=question) def create_layout_hit(self, params): """ creates a HIT for a question using the supplied layout id :param params a dict of the HIT parameters, must contain a "hit_layout" parameters with the layout id, and a "layout_params" parameter that's the dict of parameters to feed to the layout. """ # create the LayoutParameters object from the supplied params layout_params = LayoutParameters([ LayoutParameter(name, value) for name, value in params["layout_params"] ]) return self._create_hit(params, hit_layout=params["hit_layout"], layout_params=layout_params) def delete_all_hits(self): """ Permanently disables/ deletes all of the user's active HITs. :param mturk_connection: active mturk connection established by user in the notebook. :return: """ my_hits = list(self.get_all_hits()) for hit in my_hits: self.connection.disable_hit(hit.HITId) def get_assignments_object_list(self, assignment_dict): """ Returns a list of "<boto.mturk.connection.Assignment object at...>" objects assignment_dict: a dictionary of HITId-assignment object pairs """ assignments = [] for entry in assignment_dict: for assignment_object in assignment_dict[entry]: assignments.append(assignment_object) return assignments def get_results_dict(self, HIT_assignments): """ Takes a list of HIT assignment objects as input. Returns a list of dictionaries of HITs containing: HIT_id: the HIT ID worker_id: the worker ID of the Turker who completed the HIT answers: a dictionary of qid-answer field value pairs """ assignment_results = [] for assignment in HIT_assignments: HIT_dict = {} HIT_dict["assignment_object"] = assignment HIT_dict["worker_Id"] = assignment.WorkerId HIT_dict["HIT_id"] = assignment.HITId answers_dict = {} for answer in assignment.answers[0]: answers_dict[answer.qid] = answer.fields HIT_dict["answers"] = answers_dict assignment_results.append(HIT_dict) return assignment_results def get_all_results(self, hits): all_results = {} for hid, assignments in self.get_assignments(hits).items(): all_results[hid] = self.get_results_dict(assignments) return all_results def get_reviewable_hits(self, annotations=None, detailed=False): """ Get all the reviewable HITs. By default returns minimal HIT objects, but will return detailed ones (by necessity) if annotations is specified or if detailed is True :param annotations an optional set of annotations to retrieve HITs for :param detailed do you want detailed HIT objects or minimal ones :return a list of HIT objects """ minimal_hits = [] page_num = 1 while True: more_hits = self.connection.get_reviewable_hits( page_size=100, page_number=page_num) if more_hits: minimal_hits.extend(more_hits) page_num += 1 else: break if detailed or annotations is not None: detailed_hits = [ self.connection.get_hit(hit.HITId, response_groups=('Minimal', 'HITDetail')) for hit in minimal_hits ] return [ hit for hit in detailed_hits if annotation_filter(annotations, hit) ] else: return minimal_hits def get_all_hits(self, annotations=None): """ Get all the HITs. :param annotations a set of annotations to get HITs for, all HITs if not specified :return a list of HIT objects """ return [ hit for hit in self.connection.get_all_hits() if annotation_filter(annotations, hit) ] def get_assignments(self, hits=None, hit_ids=None, status=None): """ Retrieves individual assignments associated with the supplied HITs :param hits the HITs to get assignments for :status HIT status to filter by :return dict from HITId to lists of assignments """ if hit_ids is None: hit_ids = [hit.HITId for hit in hits] return { hit_id: self.connection.get_assignments(hit_id, status=status) for hit_id in hit_ids } def disable_hit(self, hit=None, hit_id=None): """ disable the specified hit (or the hit with the specified id). must specify either `hit` or `hit_id` :param hit a HIT object to disable :param hit_id a HITId to disable """ hit_id = hit.HITId if hit is not None else hit_id return self.connection.disable_hit(hit_id) def approve_assignment(self, assignment=None, assignment_id=None, feedback=None): """ approve the specified assignment (or the assigment with the specified id) must specify either `assignment` or `assignment_id` :param assignment an assignment object to approve :param assignment_id an AssignmentId to approve :param feedback optional feedback for the worker """ assignment_id = assignment.AssignmentId if assignment is not None else assignment_id return self.connection.approve_assignment(assignment_id, feedback) def reject_assignment(self, assignment=None, assignment_id=None, feedback=None): """ reject the specified assignment (or the assigment with the specified id) must specify either `assignment` or `assignment_id` :param assignment an assignment object to reject :param assignment_id an AssignmentId to reject :param feedback optional feedback for the worker """ assignment_id = assignment.AssignmentId if assignment is not None else assignment_id return self.connection.reject_assignment(assignment_id, feedback)
class MTurkProvider(object): description = 'This is a task authored by a requester on Daemo, a research crowdsourcing platform. ' \ 'Mechanical Turk workers are welcome to do it' keywords = ['daemo'] countries = ['US', 'CA'] min_hits = 1000 def __init__(self, host, aws_access_key_id, aws_secret_access_key): self.host = host self.connection = MTurkConnection( aws_access_key_id=aws_access_key_id, aws_secret_access_key=aws_secret_access_key, host=settings.MTURK_HOST) self.connection.APIVersion = "2014-08-15" if not self.host: raise ValueError("Please provide a host url") def get_connection(self): return self.connection @staticmethod def _mturk_system_qualifications(qualification): requirements = [] for item in qualification.items.all(): if item.expression['attribute'] not in [ 'location', 'approval_rate', 'total_tasks' ]: continue requirement = None if item.expression['attribute'] == 'location': op = OP_IN if item.expression['operator'] == 'in' else OP_NOT_IN requirement = MultiLocaleRequirement(op, [ val.strip() for val in item.expression['value'] if val is not None and val != '' ]) elif item.expression['attribute'] == 'approval_rate': op = OP_GT if item.expression['operator'] == 'gt' else OP_LT requirement = PercentAssignmentsApprovedRequirement( op, item.expression['value']) elif item.expression['attribute'] == 'total_tasks': op = OP_GT if item.expression['operator'] == 'gt' else OP_LT requirement = NumberHitsApprovedRequirement( op, item.expression['value']) requirements.append(requirement) return requirements def get_qualifications(self, project, boomerang_threshold, add_boomerang): requirements = [] if project.qualification is not None: requirements += self._mturk_system_qualifications( project.qualification) boomerang_qual, success = self.create_qualification_type( owner_id=project.owner_id, project_id=project.group_id, name='Boomerang Score #{}'.format(project.group_id), flag=FLAG_Q_BOOMERANG, description='No description available') boomerang = None if boomerang_threshold <= int(settings.BOOMERANG_MIDPOINT * 100): for i, bucket in enumerate(WAIT_LIST_BUCKETS): if int(bucket[1] * 100) <= boomerang_threshold: boomerang_blacklist, success = \ self.create_qualification_type(owner_id=project.owner_id, name='Boomerang Waitlist #{}-{}'.format(project.group_id, len( WAIT_LIST_BUCKETS) - i), flag=FLAG_Q_BOOMERANG, description='No description available', deny=True, project_id=project.group_id, bucket=bucket) if success and add_boomerang: boomerang = BoomerangRequirement( qualification_type_id=boomerang_blacklist.type_id, comparator=OP_DNE, integer_value=None) requirements.append(boomerang) else: boomerang = BoomerangRequirement( qualification_type_id=boomerang_qual.type_id, comparator=OP_GTEQ, integer_value=boomerang_threshold) if success and add_boomerang: requirements.append(boomerang) return Qualifications(requirements), boomerang_qual def create_hits(self, project, tasks=None, repetition=None): # if project.min_rating > 0: # return 'NOOP' if not tasks: cursor = connection.cursor() # noinspection SqlResolve query = ''' SELECT max(id) id, repetition, group_id, repetition - sum(existing_assignments) remaining_assignments, min_rating FROM ( SELECT t_rev.id, t.group_id, t.min_rating, p.repetition, CASE WHEN ma.id IS NULL OR ma.status IN (%(skipped)s, %(rejected)s, %(expired)s) THEN 0 ELSE 1 END existing_assignments FROM crowdsourcing_task t INNER JOIN crowdsourcing_project p ON t.project_id = p.id INNER JOIN crowdsourcing_task t_rev ON t_rev.group_id = t.group_id LEFT OUTER JOIN mturk_mturkhit mh ON mh.task_id = t_rev.id LEFT OUTER JOIN mturk_mturkassignment ma ON ma.hit_id = mh.id WHERE t.project_id = (%(project_id)s) AND t_rev.exclude_at IS NULL AND t_rev.deleted_at IS NULL ) t GROUP BY group_id, repetition, min_rating HAVING sum(existing_assignments) < repetition; ''' cursor.execute( query, { 'skipped': TaskWorker.STATUS_SKIPPED, 'rejected': TaskWorker.STATUS_REJECTED, 'expired': TaskWorker.STATUS_EXPIRED, 'project_id': project.id }) tasks = cursor.fetchall() rated_workers = Rating.objects.filter( origin_type=Rating.RATING_REQUESTER).count() add_boomerang = rated_workers > 0 duration = project.timeout if project.timeout is not None else datetime.timedelta( hours=24) lifetime = project.deadline - timezone.now( ) if project.deadline is not None else datetime.timedelta(days=7) for task in tasks: question = self.create_external_question(task[0]) mturk_hit = MTurkHIT.objects.filter(task_id=task[0]).first() qualifications, boomerang_qual = self.get_qualifications( project=project, boomerang_threshold=int(round(task[4], 2) * 100), add_boomerang=add_boomerang) qualifications_mask = 0 if qualifications is not None: qualifications_mask = FLAG_Q_LOCALE + FLAG_Q_HITS + FLAG_Q_RATE + FLAG_Q_BOOMERANG hit_type, success = self.create_hit_type( title=project.name, description=self.description, price=project.price, duration=duration, keywords=self.keywords, approval_delay=datetime.timedelta(days=2), qual_req=qualifications, qualifications_mask=qualifications_mask, boomerang_threshold=int(round(task[4], 2) * 100), owner_id=project.owner_id, boomerang_qual=boomerang_qual) if not success: return 'FAILURE' if mturk_hit is None: try: hit = self.connection.create_hit( hit_type=hit_type.string_id, max_assignments=task[3], lifetime=lifetime, question=question)[0] self.set_notification(hit_type_id=hit.HITTypeId) mturk_hit = MTurkHIT(hit_id=hit.HITId, hit_type=hit_type, task_id=task[0]) except MTurkRequestError as e: error = e.errors[0][0] if error == 'AWS.MechanicalTurk.InsufficientFunds': message = { "type": "ERROR", "detail": "Insufficient funds on your Mechanical Turk account!", "code": error } redis_publisher = RedisPublisher(facility='bot', users=[project.owner]) message = RedisMessage(json.dumps(message)) redis_publisher.publish_message(message) return 'FAILED' else: if mturk_hit.hit_type_id != hit_type.id: result, success = self.change_hit_type_of_hit( hit_id=mturk_hit.hit_id, hit_type_id=hit_type.string_id) if success: mturk_hit.hit_type = hit_type mturk_hit.save() return 'SUCCESS' def create_hit_type(self, owner_id, title, description, price, duration, boomerang_threshold, keywords=None, approval_delay=None, qual_req=None, qualifications_mask=0, boomerang_qual=None): hit_type = MTurkHITType.objects.filter( owner_id=owner_id, name=title, description=description, price=Decimal(str(price)), duration=duration, qualifications_mask=qualifications_mask, boomerang_threshold=boomerang_threshold).first() if hit_type is not None: return hit_type, True reward = Price(price) try: mturk_ht = self.connection.register_hit_type( title=title, description=description, reward=reward, duration=duration, keywords=keywords, approval_delay=approval_delay, qual_req=qual_req)[0] hit_type = MTurkHITType(owner_id=owner_id, name=title, description=description, price=Decimal(str(price)), keywords=keywords, duration=duration, qualifications_mask=qualifications_mask, boomerang_qualification=boomerang_qual, boomerang_threshold=boomerang_threshold) hit_type.string_id = mturk_ht.HITTypeId hit_type.save() except MTurkRequestError: return None, False return hit_type, True def create_external_question(self, task, frame_height=800): task_hash = Hashids(salt=settings.SECRET_KEY, min_length=settings.ID_HASH_MIN_LENGTH) task_id = task_hash.encode(task) url = self.host + '/mturk/task/?taskId=' + task_id question = ExternalQuestion(external_url=url, frame_height=frame_height) return question def update_max_assignments(self, task): task = Task.objects.get(id=task['id']) mturk_hit = task.mturk_hit if not mturk_hit: raise MTurkHIT.DoesNotExist( "This task is not associated to any mturk hit") assignments_completed = task.task_workers.filter(~Q(status__in=[ TaskWorker.STATUS_REJECTED, TaskWorker.STATUS_SKIPPED, TaskWorker.STATUS_EXPIRED ])).count() remaining_assignments = task.project.repetition - assignments_completed if remaining_assignments > 0 and mturk_hit.num_assignments == mturk_hit.mturk_assignments. \ filter(status=TaskWorker.STATUS_SUBMITTED).count() and \ mturk_hit.mturk_assignments.filter(status=TaskWorker.STATUS_IN_PROGRESS).count() == 0: self.add_assignments(hit_id=mturk_hit.hit_id, increment=1) self.extend_hit(hit_id=mturk_hit.hit_id) mturk_hit.status = MTurkHIT.STATUS_IN_PROGRESS mturk_hit.num_assignments += 1 mturk_hit.save() elif remaining_assignments == 0: self.expire_hit(hit_id=mturk_hit.hit_id) mturk_hit.status = MTurkHIT.STATUS_EXPIRED mturk_hit.save() elif remaining_assignments > 0 and \ mturk_hit.status == MTurkHIT.STATUS_EXPIRED: self.extend_hit(hit_id=mturk_hit.hit_id) mturk_hit.status = MTurkHIT.STATUS_IN_PROGRESS return 'SUCCESS' def get_assignment(self, assignment_id): try: return self.connection.get_assignment(assignment_id)[0], True except MTurkRequestError as e: error = e.errors[0][0] if error == 'AWS.MechanicalTurk.InvalidAssignmentState': return assignment_id, False return None, False def set_notification(self, hit_type_id): self.connection.set_rest_notification( hit_type=hit_type_id, url=self.host + '/api/mturk/notification', event_types=[ 'AssignmentReturned', 'AssignmentAbandoned', 'AssignmentAccepted', 'AssignmentSubmitted' ]) def approve_assignment(self, task_worker): task_worker_obj = TaskWorker.objects.get(id=task_worker['id']) if hasattr(task_worker_obj, 'mturk_assignments' ) and task_worker_obj.mturk_assignments.first() is not None: try: self.connection.approve_assignment( task_worker_obj.mturk_assignments.first().assignment_id) except MTurkRequestError: return False return True def reject_assignment(self, task_worker): task_worker_obj = TaskWorker.objects.get(id=task_worker['id']) if hasattr(task_worker_obj, 'mturk_assignments' ) and task_worker_obj.mturk_assignments.first() is not None: try: self.connection.reject_assignment( task_worker_obj.mturk_assignments.first().assignment_id) except MTurkRequestError: return False return True def expire_hit(self, hit_id): try: self.connection.expire_hit(hit_id) except MTurkRequestError: return False return True def disable_hit(self, hit_id): try: self.connection.disable_hit(hit_id) except MTurkRequestError: return False return True def extend_hit(self, hit_id): try: self.connection.extend_hit(hit_id=hit_id, expiration_increment=604800) # 7 days except MTurkRequestError: return False return True def add_assignments(self, hit_id, increment=1): try: self.connection.extend_hit(hit_id=hit_id, assignments_increment=increment) except MTurkRequestError: return False return True def test_connection(self): try: return self.connection.get_account_balance()[0], True except MTurkRequestError as e: error = e.errors[0][0] if error == 'AWS.NotAuthorized': return None, False return None, False def get_account_balance(self): try: return self.connection.get_account_balance()[0] except MTurkRequestError: return None def create_qualification_type(self, owner_id, name, flag, description, project_id, auto_granted=False, auto_granted_value=None, deny=False, bucket=None): # noinspection SqlResolve query = ''' SELECT * FROM ( SELECT task.target_id, task.username, round(task.task_w_avg::NUMERIC, 2) rating --round(coalesce(task.task_w_avg, requester.requester_w_avg, -- platform.platform_w_avg)::NUMERIC, 2) rating FROM ( SELECT target_id, origin_id, project_id, username, sum(weight * power((%(BOOMERANG_TASK_ALPHA)s), t.row_number)) / sum(power((%(BOOMERANG_TASK_ALPHA)s), t.row_number)) task_w_avg FROM ( SELECT r.id, r.origin_id, p.group_id project_id, weight, r.target_id, -1 + row_number() OVER (PARTITION BY target_id ORDER BY tw.created_at DESC) AS row_number, u.username username FROM crowdsourcing_rating r INNER JOIN crowdsourcing_task t ON t.id = r.task_id INNER JOIN crowdsourcing_project p ON p.id = t.project_id INNER JOIN crowdsourcing_taskworker tw ON t.id = tw.task_id AND tw.worker_id=r.target_id INNER JOIN auth_user u ON u.id = r.target_id WHERE origin_id = (%(origin_id)s) AND origin_type = (%(origin_type)s)) t GROUP BY origin_id, target_id, project_id, username) task WHERE task.project_id = (%(project_id)s) ) r ''' extra_query = 'WHERE rating BETWEEN (%(lower_bound)s) AND (%(upper_bound)s);' params = { 'origin_type': Rating.RATING_REQUESTER, 'origin_id': owner_id, 'project_id': project_id, 'BOOMERANG_REQUESTER_ALPHA': settings.BOOMERANG_REQUESTER_ALPHA, 'BOOMERANG_PLATFORM_ALPHA': settings.BOOMERANG_PLATFORM_ALPHA, 'BOOMERANG_TASK_ALPHA': settings.BOOMERANG_TASK_ALPHA } obj_params = {'upper_bound': 300, 'lower_bound': 100} if deny and bucket is not None: query += extra_query params.update({'upper_bound': bucket[1], 'lower_bound': bucket[0]}) obj_params.update({ 'upper_bound': bucket[1] * 100, 'lower_bound': bucket[0] * 100, 'is_blacklist': True }) cursor = connection.cursor() cursor.execute(query, params=params) worker_ratings_raw = cursor.fetchall() worker_ratings = [{ "worker_id": r[0], "worker_username": r[1], "rating": r[2] } for r in worker_ratings_raw] qualification = MTurkQualification.objects.filter(owner_id=owner_id, flag=flag, name=name).first() assigned_workers = [] if qualification is None: try: qualification_type = self.connection. \ create_qualification_type(name=name, description=description, status='Active', auto_granted=auto_granted, auto_granted_value=auto_granted_value)[0] qualification = MTurkQualification.objects.create( owner_id=owner_id, flag=flag, name=name, description=description, auto_granted=auto_granted, auto_granted_value=auto_granted_value, type_id=qualification_type.QualificationTypeId, **obj_params) except MTurkRequestError: return None, False else: assigned_workers = MTurkWorkerQualification.objects.values( 'worker').filter(qualification=qualification).values_list( 'worker', flat=True) for rating in worker_ratings: user_name = rating["worker_username"].split('.') if len(user_name) == 2 and user_name[0] == 'mturk': mturk_worker_id = user_name[1].upper() if mturk_worker_id not in assigned_workers: self.assign_qualification( qualification_type_id=qualification.type_id, worker_id=mturk_worker_id, value=int(rating['rating'] * 100)) defaults = { 'qualification': qualification, 'worker': mturk_worker_id, 'score': int(rating['rating'] * 100) } MTurkWorkerQualification.objects.update_or_create( qualification=qualification, worker=mturk_worker_id, defaults=defaults) return qualification, True def change_hit_type_of_hit(self, hit_id, hit_type_id): try: result = self.connection.change_hit_type_of_hit( hit_id=hit_id, hit_type=hit_type_id) except MTurkRequestError: return None, False return result, True def update_worker_boomerang(self, project_id, worker_id, task_avg, requester_avg): """ Update boomerang for project Args: project_id: worker_id: task_avg: requester_avg Returns: str """ hit = MTurkHIT.objects.select_related( 'hit_type__boomerang_qualification').filter( task__project__group_id=project_id).first() if hit is not None: qualification = hit.hit_type.boomerang_qualification worker_qual = MTurkWorkerQualification.objects.filter( qualification=qualification, worker=worker_id).first() if worker_qual is not None: self.update_score(worker_qual, score=int(task_avg * 100), override=True) else: MTurkWorkerQualification.objects.create( qualification=qualification, worker=worker_id, score=int(task_avg * 100), overwritten=True) self.assign_qualification( qualification_type_id=qualification.type_id, worker_id=worker_id, value=int(task_avg * 100)) # other_quals = MTurkWorkerQualification.objects.filter(~Q(qualification=qualification), # worker=worker_id, # overwritten=False) # for q in other_quals: # self.update_score(q, score=int(requester_avg * 100)) return 'SUCCESS' def update_score(self, worker_qual, score, override=False): if worker_qual is None: return False try: self.connection.update_qualification_score( worker_qual.qualification.type_id, worker_qual.worker, score) worker_qual.overwritten = override worker_qual.score = score worker_qual.save() except MTurkRequestError: return False return True def assign_qualification(self, qualification_type_id, worker_id, value=1): """ Revoke a qualification from a WorkerId Args: qualification_type_id: worker_id: value Returns: bool """ try: self.connection.assign_qualification(qualification_type_id, worker_id, value, send_notification=False) return True except MTurkRequestError: return False def revoke_qualification(self, qualification_type_id, worker_id): try: self.connection.revoke_qualification( qualification_type_id=qualification_type_id, subject_id=worker_id) return True except MTurkRequestError: return False def notify_workers(self, worker_ids, subject, message_text): try: self.connection.notify_workers(worker_ids, subject, message_text) return True except MTurkRequestError: return False
#!/bin/python """ Very simple example of using boto API to get your account balance. This code is based on/shamelessly stone from the tutorial at http://www.toforge.com/2011/04/boto-mturk-tutorial-create-hits/ """ from boto.mturk.connection import MTurkConnection #These keys are unique to your account, and can be found on Amazon Web Services under 'My Account' -> 'Security Credentials' ACCESS_ID = 'YOUR KEY' SECRET_KEY = 'YOUR KEY' HOST = 'mechanicalturk.sandbox.amazonaws.com' #Set up a connection with MTurk conn = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) print 'Your current balance: ', conn.get_account_balance()
class MTurkServices(object): ''' MTurk services ''' def __init__(self, aws_access_key_id, aws_secret_access_key, is_sandbox): self.update_credentials(aws_access_key_id, aws_secret_access_key) self.set_sandbox(is_sandbox) self.valid_login = self.verify_aws_login() if not self.valid_login: print 'WARNING *****************************' print 'Sorry, AWS Credentials invalid.\nYou will only be able to '\ 'test experiments locally until you enter\nvalid '\ 'credentials in the AWS Access section of ~/.psiturkconfig\n' def update_credentials(self, aws_access_key_id, aws_secret_access_key): ''' Update credentials ''' self.aws_access_key_id = aws_access_key_id self.aws_secret_access_key = aws_secret_access_key def set_sandbox(self, is_sandbox): ''' Set sandbox ''' self.is_sandbox = is_sandbox def get_reviewable_hits(self): ''' Get reviewable HITs ''' if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False reviewable_hits = [hit for hit in hits if hit.HITStatus == "Reviewable" \ or hit.HITStatus == "Reviewing"] hits_data = [MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration }) for hit in reviewable_hits] return hits_data def get_all_hits(self): ''' Get all HITs ''' if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False hits_data = [MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in hits] return hits_data def get_active_hits(self): ''' Get active HITs ''' if not self.connect_to_turk(): return False # hits = self.mtc.search_hits() try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False active_hits = [hit for hit in hits if not hit.expired] hits_data = [MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in active_hits] return hits_data def get_workers(self, assignment_status=None, chosen_hit=None): ''' Get workers ''' if not self.connect_to_turk(): return False try: if chosen_hit: hit_ids = [chosen_hit] else: hits = self.mtc.get_all_hits() hit_ids = [hit.HITId for hit in hits] workers_nested = [] page_size=100 for hit_id in hit_ids: current_page_number=1 hit_assignments = self.mtc.get_assignments( hit_id, status=assignment_status, sort_by='SubmitTime', page_size=page_size, page_number=current_page_number ) totalNumResults = int(hit_assignments.TotalNumResults) total_pages = (totalNumResults // page_size) + (totalNumResults % page_size > 0) #do integer division then round up if necessary while current_page_number < total_pages: current_page_number += 1 hit_assignments += self.mtc.get_assignments( hit_id, status=assignment_status, sort_by='SubmitTime', page_size=page_size, page_number=current_page_number ) workers_nested.append(hit_assignments) workers = [val for subl in workers_nested for val in subl] # Flatten nested lists except MTurkRequestError: return False worker_data = [{ 'hitId': worker.HITId, 'assignmentId': worker.AssignmentId, 'workerId': worker.WorkerId, 'submit_time': worker.SubmitTime, 'accept_time': worker.AcceptTime, 'status': worker.AssignmentStatus } for worker in workers] return worker_data def get_worker(self, assignment_id): if not self.connect_to_turk(): return False try: worker = self.mtc.get_assignment(assignment_id)[0] except MTurkRequestError as e: return False worker_data = [{ 'hitId': worker.HITId, 'assignmentId': worker.AssignmentId, 'workerId': worker.WorkerId, 'submit_time': worker.SubmitTime, 'accept_time': worker.AcceptTime, 'status': worker.AssignmentStatus }] return worker_data def bonus_worker(self, assignment_id, amount, reason=""): ''' Bonus worker ''' if not self.connect_to_turk(): return False try: bonus = MTurkConnection.get_price_as_price(amount) assignment = self.mtc.get_assignment(assignment_id)[0] worker_id = assignment.WorkerId self.mtc.grant_bonus(worker_id, assignment_id, bonus, reason) return True except MTurkRequestError as exception: print exception return False def approve_worker(self, assignment_id): ''' Approve worker ''' if not self.connect_to_turk(): return False try: self.mtc.approve_assignment(assignment_id, feedback=None) return True except MTurkRequestError as e: return False def reject_worker(self, assignment_id): ''' Reject worker ''' if not self.connect_to_turk(): return False try: self.mtc.reject_assignment(assignment_id, feedback=None) return True except MTurkRequestError: return False def unreject_worker(self, assignment_id): ''' Unreject worker ''' if not self.connect_to_turk(): return False try: self.mtc.approve_rejected_assignment(assignment_id) return True except MTurkRequestError: return False def verify_aws_login(self): ''' Verify AWS login ''' if ((self.aws_access_key_id == 'YourAccessKeyId') or (self.aws_secret_access_key == 'YourSecretAccessKey')): return False else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict( aws_access_key_id=self.aws_access_key_id, aws_secret_access_key=self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**mturkparams) try: self.mtc.get_account_balance() except MTurkRequestError as exception: print exception.error_message return False else: return True def connect_to_turk(self): ''' Connect to turk ''' if not self.valid_login: print 'Sorry, unable to connect to Amazon Mechanical Turk. AWS '\ 'credentials invalid.' return False if self.is_sandbox: host = 'mechanicalturk.sandbox.amazonaws.com' else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict( aws_access_key_id=self.aws_access_key_id, aws_secret_access_key=self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**mturkparams) return True def configure_hit(self, hit_config): ''' Configure HIT ''' # configure question_url based on the id experiment_portal_url = hit_config['ad_location'] frame_height = 600 mturk_question = ExternalQuestion(experiment_portal_url, frame_height) # Qualification: quals = Qualifications() approve_requirement = hit_config['approve_requirement'] quals.add( PercentAssignmentsApprovedRequirement("GreaterThanOrEqualTo", approve_requirement)) number_hits_approved = hit_config['number_hits_approved'] quals.add( NumberHitsApprovedRequirement("GreaterThanOrEqualTo", number_hits_approved)) require_master_workers = hit_config['require_master_workers'] if require_master_workers: quals.add(MasterRequirement(sandbox=self.is_sandbox)) if hit_config['us_only']: quals.add(LocaleRequirement("EqualTo", "US")) # Create a HIT type for this HIT. hit_type = self.mtc.register_hit_type( hit_config['title'], hit_config['description'], hit_config['reward'], hit_config['duration'], keywords=hit_config['keywords'], approval_delay=None, qual_req=quals)[0] # Check the config file to see if notifications are wanted. config = PsiturkConfig() config.load_config() try: url = config.get('Server Parameters', 'notification_url') all_event_types = [ "AssignmentAccepted", "AssignmentAbandoned", "AssignmentReturned", "AssignmentSubmitted", "HITReviewable", "HITExpired", ] self.mtc.set_rest_notification( hit_type.HITTypeId, url, event_types=all_event_types) except: pass # Specify all the HIT parameters self.param_dict = dict( hit_type=hit_type.HITTypeId, question=mturk_question, lifetime=hit_config['lifetime'], max_assignments=hit_config['max_assignments'], questions=None, response_groups=[ 'Minimal', 'HITDetail', 'HITQuestion', 'HITAssignmentSummary' ]) def check_balance(self): ''' Check balance ''' if not self.connect_to_turk(): return '-' return self.mtc.get_account_balance()[0] # TODO (if valid AWS credentials haven't been provided then # connect_to_turk() will fail, not error checking here and elsewhere) def create_hit(self, hit_config): ''' Create HIT ''' try: if not self.connect_to_turk(): return False self.configure_hit(hit_config) myhit = self.mtc.create_hit(**self.param_dict)[0] self.hitid = myhit.HITId except MTurkRequestError as e: print e return False else: return self.hitid # TODO(Jay): Have a wrapper around functions that serializes them. # Default output should not be serialized. def expire_hit(self, hitid): ''' Expire HIT ''' if not self.connect_to_turk(): return False try: self.mtc.expire_hit(hitid) return True except MTurkRequestError: print "Failed to expire HIT. Please check the ID and try again." return False def dispose_hit(self, hitid): ''' Dispose HIT ''' if not self.connect_to_turk(): return False try: self.mtc.dispose_hit(hitid) except Exception, e: print "Failed to dispose of HIT %s. Make sure there are no "\ "assignments remaining to be reviewed." % hitid
def createhitsubmit(request): ''' Submitting mkturk to amazon web service. @param request: Django http request containing keys and other task arguments. ''' from boto.mturk.connection import MTurkConnection from boto.mturk.question import QuestionContent,Question,QuestionForm,Overview,AnswerSpecification,SelectionAnswer,FormattedContent,FreeTextAnswer print 'Ok' # Get request data from the front-end requestJson = json.loads(request.body) user_aws_secret_key = requestJson['aws_secret_key'] user_aws_access_key_id = requestJson['aws_access_key_id'] task_selected_docs = requestJson['task_selected_docs'] #id task_title = requestJson['task_title'] #task_dataset = requestJson['task_dataset'] # id task_description = requestJson['task_description'] task_duration = requestJson['task_duration'] task_max_assignment = requestJson['task_max_assignment'] task_reward = requestJson['task_reward'] # adjust host setting, depending on whether HIT is live (production) or in testing mode (sandbox) mode = "sandbox" #mode ="production" if mode=="production": HOST='mechanicalturk.amazonaws.com' else: HOST='mechanicalturk.sandbox.amazonaws.com' mtc = MTurkConnection(aws_access_key_id= user_aws_access_key_id, aws_secret_access_key= user_aws_secret_key, host=HOST) overview = Overview() overview.append_field('Title', task_title) overview.append(FormattedContent('<b>' + task_description + '</b><p></p>')) tableStr = '<ul>' for docID in task_selected_docs: docText = Doc.objects.get(pk = docID) tableStr += '<li>' + docText.text + '</li>' tableStr += '</ul>' overview.append(FormattedContent(tableStr)) qc2 = QuestionContent() qc2.append_field('Title','What do you find?') fta2 = FreeTextAnswer() q2 = Question(identifier="comments", content=qc2, answer_spec=AnswerSpecification(fta2)) #--------------- BUILD THE QUESTION FORM ------------------- question_form = QuestionForm() question_form.append(overview) question_form.append(q2) print 'Before create hit' #--------------- CREATE THE HIT ------------------- try: creathitReturnValue = mtc.create_hit(questions=question_form, max_assignments= task_max_assignment, title=task_title, description=task_description, keywords='SomeKeywords', duration = task_duration, reward= task_reward) except Exception as e: print e print 'after crate hit' return HttpResponse(json.dumps({'data' : mtc.get_account_balance()}), content_type = "application/json")
class MyMTurk: AWS_ACCESS_KEY_ID = 'skip' AWS_SECRET_ACCESS_KEY = 'skip' HOST_SANDBOX = 'mechanicalturk.sandbox.amazonaws.com' HOST_MTURK = 'mechanicalturk.amazonaws.com' EXTERNAL_URL = 'http://redbug0314.blogspot.com/p/imcrowd.html' def __init__( self ): #connect to MTurk self.connect = MTurkConnection( self.AWS_ACCESS_KEY_ID, self.AWS_SECRET_ACCESS_KEY, host=self.HOST_SANDBOX ) #Qualification setting q = self.qualifications = Qualifications() # if required_to_preview == True unqualified user even can't view the hit. # q.add( PercentAssignmentsApprovedRequirement( comparator="GreaterThan", integer_value="95" ) ) q.add( AdultRequirement( comparator="EqualTo", integer_value="1" ) ) def register_hit_type( self ): try: reg_hit_type = self.connect.register_hit_type( title="Nine Picture!", description="Choose some best pictures which you think is the best from following pictures.", reward=0.01, duration=60 * 30, keywords="steak, photo", approval_delay=datetime.timedelta( days=1 ), qual_req=self.qualifications ) except MTurkRequestError as e: print "register hit type error:\n status: %s reason: %s\n body: %s" % ( e.status, e.reason, e.body ) else: self.hit_type_id = reg_hit_type print "hit type id %s" % reg_hit_type def question_form( self ): qc = QuestionContent() # qc.append_field( 'Title', 'Is she hot?' ) qc.append( Binary( 'image', 'jpg', 'http://www.miranchomeatmarket.com/images/T-%20bone%20steak.jpg', 'steak' ) ) q = Question( identifier="This is the first girl!", content=qc, answer_spec=AnswerSpecification( FreeTextAnswer() ), is_required=True, display_name="This is display name" ) qf = QuestionForm() qf.append( q ) if self.hit_type_id: try: create_hit_rs = self.connect.create_hit( hit_type=self.hit_type_id, question=qf, lifetime=datetime.timedelta( days=14 ), max_assignments=10, annotation="This is a annotation" ) except MTurkRequestError as e: print "create hit type error:\n status: %s reason: %s\n body: %s" % ( e.status, e.reason, e.body ) else: print "success!! key: %s" % create_hit_rs def question_form_formatted_content( self ): qc = QuestionContent() formatted_xhtml = """\ <table border="1"> <tr> <td></td> <td align="center">1</td> <td align="center">2</td> <td align="center">3</td> </tr> <tr> <td align="right">A</td> <td align="center"><b>X</b></td> <td align="center"> </td> <td align="center"><b>O</b></td> </tr> <tr> <td align="right">B</td> <td align="center"> </td> <td align="center"><b>O</b></td> <td align="center"> </td> </tr> <tr> <td align="right">C</td> <td align="center"> </td> <td align="center"> </td> <td align="center"><b>X</b></td> </tr> <tr> <td align="center" colspan="4">It is <b>X</b>'s turn.</td> </tr> </table> """ qc.append( FormattedContent( formatted_xhtml ) ) q = Question( identifier="Formatted content test!", content=qc, answer_spec=AnswerSpecification( SelectionAnswer( min=1, max=5, style='checkbox', selections=[ ( Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcSh1HXq3WyOvvG7-AgvNugKC2LzImMUvUDNTuDAPwVKuw8NZzvLN62pGYhX:farm1.static.flickr.com/21/24204504_e143536a2e.jpg', 'steak1' ).get_as_xml(), 'img1' ), ( Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcTkMoChevUBvQfmfksKDBM5oj4V2ruj6riqv7kC-_6qf9MR0igeBlJLkSI:www.miranchomeatmarket.com/images/T-%2520bone%2520steak.jpg', 'steak2' ).get_as_xml(), 'img2' ), ( Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcSttsqT7kj9siDKZg1p4fU6W9IFlMZHCFSxFd49ECJR1Bu_1QlHQwmH1DU:img4.myrecipes.com/i/recipes/ck/06/08/grilled-steak-ck-1215910-l.jpg', 'steak3' ).get_as_xml(), 'img3' ), ( Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcRfdQ-vuNt-W4W7JZRkAmbZpE6LLA0puCQs5erSzrGtsOY8H8t-vgEzqA:www.greendiamondgrille.com/images/new/NewYorkStripSteak.jpg', 'steak4' ).get_as_xml(), 'img4' ), ( Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcTsJzCp6En1R9yvFQw7bGsSxiiQCqlMrFg7XCbcJ13G39Aa3e6ZilWW34oI:www.bunrab.com/dailyfeed/dailyfeed_images_jan-07/df07_01-08_steak.jpg', 'steak5' ).get_as_xml(), 'img5' ), ( Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcTkMoChevUBvQfmfksKDBM5oj4V2ruj6riqv7kC-_6qf9MR0igeBlJLkSI:www.miranchomeatmarket.com/images/T-%2520bone%2520steak.jpg', 'steak2' ).get_as_xml(), 'img6' ), ( Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcSttsqT7kj9siDKZg1p4fU6W9IFlMZHCFSxFd49ECJR1Bu_1QlHQwmH1DU:img4.myrecipes.com/i/recipes/ck/06/08/grilled-steak-ck-1215910-l.jpg', 'steak3' ).get_as_xml(), 'img7' ), ( Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcRfdQ-vuNt-W4W7JZRkAmbZpE6LLA0puCQs5erSzrGtsOY8H8t-vgEzqA:www.greendiamondgrille.com/images/new/NewYorkStripSteak.jpg', 'steak4' ).get_as_xml(), 'img8' ), ( Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcTsJzCp6En1R9yvFQw7bGsSxiiQCqlMrFg7XCbcJ13G39Aa3e6ZilWW34oI:www.bunrab.com/dailyfeed/dailyfeed_images_jan-07/df07_01-08_steak.jpg', 'steak5' ).get_as_xml(), 'img9' ) ], type='binary' ) ), is_required=True, display_name="This is display name" ) qf = QuestionForm() qf.append( q ) if self.hit_type_id: try: create_hit_rs = self.connect.create_hit( hit_type=self.hit_type_id, question=qf, lifetime=datetime.timedelta( days=14 ), max_assignments=1, annotation="This is a annotation" ) except MTurkRequestError as e: print "create hit type error:\n status: %s reason: %s\n body: %s" % ( e.status, e.reason, e.body ) else: print "success!! key: %s" % create_hit_rs def external_question( self ): q = ExternalQuestion( external_url="http://www.kernel.org/pub/software/scm/git/docs/everyday.html", frame_height=200 ) # keywords = ['image', 'filter', 'google'] # #create hit without id # create_hit_rs = self.connect.create_hit( question=q, lifetime=60 * 65, max_assignments=2, title="Google Image Filter", keywords=keywords, reward=0.05, duration=60 * 6, approval_delay=60 * 60, annotation='An annotation from boto external question test', response_groups=['Minimal', 'HITDetail', 'HITQuestion', 'HITAssignmentSummary', ], qualifications=self.qualifications ) #create hit with id if self.hit_type_id: try: hit = self.connect.create_hit( hit_type=self.hit_type_id, question=q, lifetime=datetime.timedelta( days=14 ), max_assignments=1, annotation="This is a annotation" ) except MTurkRequestError as e: print "register hit type error:\n status: %s reason: %s\n body: %s" % ( e.status, e.reason, e.body ) else: print "hit id: %s " % hit[0].HITId print "hit type id: %s " % hit[0].HITTypeId def get_account_balance( self ): print self.connect.get_account_balance() def getHits( self ): print self.connect.get_all_hits() def getHit( self, hit_id ): hit_rs = self.connect.get_hit( hit_id ) hit = hit_rs[0] for k, v in hit.__dict__.items(): print "%s: %s" % ( k, v ) def searchHits( self ): print self.connect.search_hits() def getAssignments( self, hit_id ): print self.connect.get_assignments( hit_id ) def getReviewableHits( self ): print self.connect.get_reviewable_hits()
def new_sugg_hit(PIN_IMAGE_URL, PIN_IMAGE_TITLE): mtc = MTurkConnection(aws_access_key_id=AWS_ACCESS_KEY_ID, aws_secret_access_key=AWS_SECRET_ACCESS_KEY, host=HOST) if debug: print mtc.get_account_balance() title = 'Match these Pictures to Macy\'s Products' description = 'Look at this photo and match it to Macy\'s products' keywords = 'clothing, rating, opinions, easy, quick, macys' #make overview overview = Overview() overview.append_field('Title', 'Find three Macys.com Product Web IDs That Match') overview.append( FormattedContent('<img src="' + PIN_IMAGE_URL + '" alt="Pintrest Image" />' '<br />' + PIN_IMAGE_TITLE)) #make webid1 qc1 = QuestionContent() qc1.append_field('Title', 'First WebID Code') fta1 = FreeTextAnswer(num_lines=1) q1 = Question(identifier="FirstWebCode", content=qc1, answer_spec=AnswerSpecification(fta1)) #make webid2 qc2 = QuestionContent() qc2.append_field('Title', 'Second WebID Code') fta2 = FreeTextAnswer(num_lines=1) q2 = Question(identifier="SecondWebCode", content=qc2, answer_spec=AnswerSpecification(fta2)) #make webid1 qc3 = QuestionContent() qc3.append_field('Title', 'Third WebID Code') fta3 = FreeTextAnswer(num_lines=1) q3 = Question(identifier="ThirdWebCode", content=qc3, answer_spec=AnswerSpecification(fta3)) #make question form question_form = QuestionForm() question_form.append(overview) question_form.append(q1) question_form.append(q2) question_form.append(q3) #--------------- CREATE THE HIT ------------------- mtc.create_hit(questions=question_form, max_assignments=1, title=title, description=description, keywords=keywords, duration=60 * 5, reward=0.05)
#!/bin/python """ Very simple example of using boto API to get your account balance. This code is based on/shamelessly stone from the tutorial at http://www.toforge.com/2011/04/boto-mturk-tutorial-create-hits/ """ from boto.mturk.connection import MTurkConnection #These keys are unique to your account, and can be found on Amazon Web Services under 'My Account' -> 'Security Credentials' ACCESS_ID = 'YOUR KEY' SECRET_KEY = 'YOUR KEY' HOST = 'mechanicalturk.sandbox.amazonaws.com' #Set up a connection with MTurk conn = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) print 'Your current balance: ', conn.get_account_balance()
class MyMTurk: AWS_ACCESS_KEY_ID = 'skip' AWS_SECRET_ACCESS_KEY = 'skip' HOST_SANDBOX = 'mechanicalturk.sandbox.amazonaws.com' HOST_MTURK = 'mechanicalturk.amazonaws.com' EXTERNAL_URL = 'http://redbug0314.blogspot.com/p/imcrowd.html' def __init__(self): #connect to MTurk self.connect = MTurkConnection(self.AWS_ACCESS_KEY_ID, self.AWS_SECRET_ACCESS_KEY, host=self.HOST_SANDBOX) #Qualification setting q = self.qualifications = Qualifications() # if required_to_preview == True unqualified user even can't view the hit. # q.add( PercentAssignmentsApprovedRequirement( comparator="GreaterThan", integer_value="95" ) ) q.add(AdultRequirement(comparator="EqualTo", integer_value="1")) def register_hit_type(self): try: reg_hit_type = self.connect.register_hit_type( title="Nine Picture!", description= "Choose some best pictures which you think is the best from following pictures.", reward=0.01, duration=60 * 30, keywords="steak, photo", approval_delay=datetime.timedelta(days=1), qual_req=self.qualifications) except MTurkRequestError as e: print "register hit type error:\n status: %s reason: %s\n body: %s" % ( e.status, e.reason, e.body) else: self.hit_type_id = reg_hit_type print "hit type id %s" % reg_hit_type def question_form(self): qc = QuestionContent() # qc.append_field( 'Title', 'Is she hot?' ) qc.append( Binary( 'image', 'jpg', 'http://www.miranchomeatmarket.com/images/T-%20bone%20steak.jpg', 'steak')) q = Question(identifier="This is the first girl!", content=qc, answer_spec=AnswerSpecification(FreeTextAnswer()), is_required=True, display_name="This is display name") qf = QuestionForm() qf.append(q) if self.hit_type_id: try: create_hit_rs = self.connect.create_hit( hit_type=self.hit_type_id, question=qf, lifetime=datetime.timedelta(days=14), max_assignments=10, annotation="This is a annotation") except MTurkRequestError as e: print "create hit type error:\n status: %s reason: %s\n body: %s" % ( e.status, e.reason, e.body) else: print "success!! key: %s" % create_hit_rs def question_form_formatted_content(self): qc = QuestionContent() formatted_xhtml = """\ <table border="1"> <tr> <td></td> <td align="center">1</td> <td align="center">2</td> <td align="center">3</td> </tr> <tr> <td align="right">A</td> <td align="center"><b>X</b></td> <td align="center"> </td> <td align="center"><b>O</b></td> </tr> <tr> <td align="right">B</td> <td align="center"> </td> <td align="center"><b>O</b></td> <td align="center"> </td> </tr> <tr> <td align="right">C</td> <td align="center"> </td> <td align="center"> </td> <td align="center"><b>X</b></td> </tr> <tr> <td align="center" colspan="4">It is <b>X</b>'s turn.</td> </tr> </table> """ qc.append(FormattedContent(formatted_xhtml)) q = Question( identifier="Formatted content test!", content=qc, answer_spec=AnswerSpecification( SelectionAnswer( min=1, max=5, style='checkbox', selections=[ (Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcSh1HXq3WyOvvG7-AgvNugKC2LzImMUvUDNTuDAPwVKuw8NZzvLN62pGYhX:farm1.static.flickr.com/21/24204504_e143536a2e.jpg', 'steak1').get_as_xml(), 'img1'), (Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcTkMoChevUBvQfmfksKDBM5oj4V2ruj6riqv7kC-_6qf9MR0igeBlJLkSI:www.miranchomeatmarket.com/images/T-%2520bone%2520steak.jpg', 'steak2').get_as_xml(), 'img2'), (Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcSttsqT7kj9siDKZg1p4fU6W9IFlMZHCFSxFd49ECJR1Bu_1QlHQwmH1DU:img4.myrecipes.com/i/recipes/ck/06/08/grilled-steak-ck-1215910-l.jpg', 'steak3').get_as_xml(), 'img3'), (Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcRfdQ-vuNt-W4W7JZRkAmbZpE6LLA0puCQs5erSzrGtsOY8H8t-vgEzqA:www.greendiamondgrille.com/images/new/NewYorkStripSteak.jpg', 'steak4').get_as_xml(), 'img4'), (Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcTsJzCp6En1R9yvFQw7bGsSxiiQCqlMrFg7XCbcJ13G39Aa3e6ZilWW34oI:www.bunrab.com/dailyfeed/dailyfeed_images_jan-07/df07_01-08_steak.jpg', 'steak5').get_as_xml(), 'img5'), (Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcTkMoChevUBvQfmfksKDBM5oj4V2ruj6riqv7kC-_6qf9MR0igeBlJLkSI:www.miranchomeatmarket.com/images/T-%2520bone%2520steak.jpg', 'steak2').get_as_xml(), 'img6'), (Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcSttsqT7kj9siDKZg1p4fU6W9IFlMZHCFSxFd49ECJR1Bu_1QlHQwmH1DU:img4.myrecipes.com/i/recipes/ck/06/08/grilled-steak-ck-1215910-l.jpg', 'steak3').get_as_xml(), 'img7'), (Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcRfdQ-vuNt-W4W7JZRkAmbZpE6LLA0puCQs5erSzrGtsOY8H8t-vgEzqA:www.greendiamondgrille.com/images/new/NewYorkStripSteak.jpg', 'steak4').get_as_xml(), 'img8'), (Binary( 'image', 'jpg', 'http://images.google.com/images?q=tbn:ANd9GcTsJzCp6En1R9yvFQw7bGsSxiiQCqlMrFg7XCbcJ13G39Aa3e6ZilWW34oI:www.bunrab.com/dailyfeed/dailyfeed_images_jan-07/df07_01-08_steak.jpg', 'steak5').get_as_xml(), 'img9') ], type='binary')), is_required=True, display_name="This is display name") qf = QuestionForm() qf.append(q) if self.hit_type_id: try: create_hit_rs = self.connect.create_hit( hit_type=self.hit_type_id, question=qf, lifetime=datetime.timedelta(days=14), max_assignments=1, annotation="This is a annotation") except MTurkRequestError as e: print "create hit type error:\n status: %s reason: %s\n body: %s" % ( e.status, e.reason, e.body) else: print "success!! key: %s" % create_hit_rs def external_question(self): q = ExternalQuestion( external_url= "http://www.kernel.org/pub/software/scm/git/docs/everyday.html", frame_height=200) # keywords = ['image', 'filter', 'google'] # #create hit without id # create_hit_rs = self.connect.create_hit( question=q, lifetime=60 * 65, max_assignments=2, title="Google Image Filter", keywords=keywords, reward=0.05, duration=60 * 6, approval_delay=60 * 60, annotation='An annotation from boto external question test', response_groups=['Minimal', 'HITDetail', 'HITQuestion', 'HITAssignmentSummary', ], qualifications=self.qualifications ) #create hit with id if self.hit_type_id: try: hit = self.connect.create_hit( hit_type=self.hit_type_id, question=q, lifetime=datetime.timedelta(days=14), max_assignments=1, annotation="This is a annotation") except MTurkRequestError as e: print "register hit type error:\n status: %s reason: %s\n body: %s" % ( e.status, e.reason, e.body) else: print "hit id: %s " % hit[0].HITId print "hit type id: %s " % hit[0].HITTypeId def get_account_balance(self): print self.connect.get_account_balance() def getHits(self): print self.connect.get_all_hits() def getHit(self, hit_id): hit_rs = self.connect.get_hit(hit_id) hit = hit_rs[0] for k, v in hit.__dict__.items(): print "%s: %s" % (k, v) def searchHits(self): print self.connect.search_hits() def getAssignments(self, hit_id): print self.connect.get_assignments(hit_id) def getReviewableHits(self): print self.connect.get_reviewable_hits()
class HaCRSTurker: def __init__(self): self.config = HaCRSUtil.get_config('../config.ini') HOST = self.config.get('mturk', 'host') AWS_ACCESS_KEY_ID = self.config.get('mturk', 'access_key_id') AWS_SECRET_ACCESS_KEY = self.config.get('mturk', 'secret_access_key') self.MTconnection = MTurkConnection( aws_access_key_id=AWS_ACCESS_KEY_ID, aws_secret_access_key=AWS_SECRET_ACCESS_KEY, host=HOST) self.db = HaCRSDB() def get_balance(self): print self.MTconnection.get_account_balance() def expire_all_hits(self): all_hits = self.MTconnection.get_all_hits() for hit in all_hits: if hit.expired: continue try: self.MTconnection.expire_hit(hit.HITId) print 'Expired HIT' except Exception as e: print 'Could not expire: {}'.format(e) def delete_all_mturk_hits(self): all_hits = self.MTconnection.get_all_hits() for hit in all_hits: print 'expire/dispose' self.MTconnection.expire_hit(hit.HITId) self.MTconnection.dispose_hit(hit.HITId) def get_all_mturk_hits(self): all_hits = self.MTconnection.get_all_hits() return all_hits # TODO: HITs available via API, but not via Amazon Web Sandbox def push_tasklet_mturk(self, keywords): sdescription = self.config.get('mturk', 'shortdescr') frame_height = self.config.get('mturk', 'frameheight') #url = "https://cgcturk.hacked.jp/tasklet/{}/".format(tasklet['id']) url = "https://cgcturk.hacked.jp/pick_tasklet/{}/".format(keywords) #keywords = tasklet['keywords'] #amount = tasklet['amount'] if keywords == 'easy': amount = 1.00 elif keywords in ['medium', 'hard', 'very_hard']: amount = 2.00 elif keywords == 'priority': amount = 4.00 else: print 'Error' sys.exit(1) questionform = ExternalQuestion(url, frame_height) title = 'HELP AN AI!!! We are students building an artificial intelligence to find bugs in programs to keep the internet safe' sdescription = 'We are students building an artificial intelligence system that finds bugs in programs and keeps the internet safe from malware. BUT IT NEEDS YOUR HELP! Play with programs to find functions that it missed, and get $$$!' hit_result = self.MTconnection.create_hit( title='[{}] {}'.format(keywords, title), description=sdescription, keywords=keywords, max_assignments=1, question=questionform, reward=Price(amount=amount), response_groups=('Minimal', 'HITDetail'), # ? ) assert len(hit_result) == 1 mturkid = self.db.create_mturk_resource(hit_result[0].HITId, hit_result[0].HITGroupId) #self.db.add_mturk_tasklet_association(tasklet['id'], mturkid) #self.db.commit() return mturkid, hit_result def push_tasks_mturk(self): frame_height = self.config.get('mturk', 'frameheight') amount = 0.01 tasklets = self.db.get_unassigned_tasklets() sdescription = self.config.get('mturk', 'shortdescr') for tasklet in tasklets: print 'pushing!' url = "https://cgcturk.hacked.jp/tasklet/{}/".format(tasklet['id']) keywords = ["easy"] questionform = ExternalQuestion(url, frame_height) hit_result = self.MTconnection.create_hit( title=HaCRSUtil.get_tasklet_name(tasklet), description=sdescription, keywords=keywords, max_assignments=1, question=questionform, reward=Price(amount=amount), response_groups=('Minimal', 'HITDetail'), # ? ) assert len(hit_result) == 1 mturkid = self.db.create_mturk_resource(hit_result[0].HITId, hit_result[0].HITGroupId) self.db.add_mturk_tasklet_association(tasklet['id'], mturkid) self.db.commit() def show_seed_tasklets(self): pprint(self.db.get_seed_tasklets()) def get_hit(self, hitid): try: hit = self.MTconnection.get_hit(hitid) except Exception as e: return None if hit != None: return hit[0] def get_assignment_from_hit(self, hitid): try: assignments = self.MTconnection.get_assignments(hitid) return assignments[0] except Exception as e: return None def get_approved_seeding_tasklets(self): for program in json.load( open(self.config.get('general', 'programsjson'))): pid = self.db.lookup_program(program) program = None approved = set() for tasklet in self.db.get_latest_seed_tasklets(): turkinfos = self.db.get_mturk_infos(tasklet['id']) try: #hit = self.MTconnection.get_hit(turkinfos['hitid']) assignments = self.MTconnection.get_assignments( turkinfos['hitid']) if len(assignments) == 0: continue if assignments[0].AssignmentStatus == 'Approved': approved.add(self.db.get_tasklet_program(tasklet['id'])) except Exception as e: #print e pass return list(approved)
class Mturk(): def __init__(self): self.config = self.set_config() self.mturk = MTurkConnection( aws_access_key_id=self.config['aws_access_key_id'], aws_secret_access_key=self.config['aws_secret_access_key'], host=self.config['host']) self.mturk_tmpl = MturkTmpl() def set_config(self, config_path="config.yml"): with open(config_path, 'r') as file: config = yaml.load(file) return config def account_balance(self): account_balance = self.mturk.get_account_balance() print("Testing connection: You have a balance of: {}".format( account_balance)) def get_hits(self): return self.mturk.get_all_hits() def get_all_assignments(self, hit_id): page_size = 100 assignments = self.mturk.get_assignments(hit_id, page_size=page_size) total_records = int(assignments.TotalNumResults) get_page_assignments = lambda page: self.mturk.get_assignments( hit_id, page_size=page_size, page_number=page) page_nums = self.mturk._get_pages(page_size=page_size, total_records=total_records) assignments_sets = itertools.imap(get_page_assignments, page_nums) return itertools.chain.from_iterable(assignments_sets) def remove_old_hits(self): # Disable old hits. for hit in self.get_hits(): print("Hit {} has been removed.".format(hit.HITId)) self.mturk.disable_hit(hit.HITId) def cal_reward(self, data): read_instruction = 3.0 word_count = len(data['ents']) * 1 / 30.0 return round((read_instruction + word_count) / 60.0 * 6.0, 2) def create_hit(self, data): # These parameters define the HIT that will be created # question is what we defined above # max_assignments is the # of unique Workers you're requesting # title, description, and keywords help Workers find your HIT # duration is the # of seconds Workers have to complete your HIT # reward is what Workers will be paid when you approve their work # Check out the documentation on CreateHIT for more details response = self.mturk.create_hit( question=self.mturk_tmpl.html_question(data), max_assignments=1, title=self.config['title'], description=self.config['description'], keywords=self.config['keywords'], duration=120, reward=self.cal_reward(data)) return response
""" import csv import argparse from boto.mturk.connection import MTurkConnection from boto.mturk.price import Price from boto.exception import AWSConnectionError, EC2ResponseError parser = argparse.ArgumentParser(description='Grant bonuses for HITs on Amazon Mechanical Turk') parser.add_argument('-experiment', required=True, help='(required) The name of the experiment you are granting bonuses for') args = parser.parse_args() try: conn = MTurkConnection(is_secure=True) mtbal = conn.get_account_balance() if len(mtbal) == 1: print "Available balance: ", mtbal[0] elif len(mtbal) == 2: print "Available balance: ", mtbal[0] print "OnHold balance: ", mtbal[1] avbal = mtbal[0] #if (bonuscost > avbal.amount): # print "Insufficient funds to pay bonuses! Add $%.2f to your account before proceeding" % (bonuscost - avbal.amount) #else: with open('bonus.' + args.expt_name + '.csv', 'r') as csvinfile: fields = ('worker', 'trials', 'bonus', 'assignment') bonuses = csv.DictReader(csvinfile, fieldnames=fields) bonuses.next() # pop off the header row for row in bonuses:
import boto from boto.mturk.connection import MTurkConnection # Create the connection to MTurk mtc = MTurkConnection(aws_access_key_id='MYACCESSKEYID', aws_secret_access_key='MYSECRETACCESSKEY', host='mechanicalturk.amazonaws.com') # Print account balance to show connection is working account_balance = mtc.get_account_balance()[0] print "You have a balance of: {}".format(account_balance) # Create the qualification response = mtc.create_qualification_type(name="NAME", description="DESCRIPTION", status="Active|Inactive", auto_granted=True|False, auto_granted_value=###)
class MTurkServices: def __init__(self, aws_access_key_id, aws_secret_access_key, is_sandbox): self.update_credentials(aws_access_key_id, aws_secret_access_key) self.set_sandbox(is_sandbox) self.validLogin = self.verify_aws_login() if not self.validLogin: print 'WARNING *****************************' print 'Sorry, AWS Credentials invalid.\nYou will only be able to '\ + 'test experiments locally until you enter\nvalid '\ + 'credentials in the AWS Access section of ~/.psiturkconfig\n' def update_credentials(self, aws_access_key_id, aws_secret_access_key): self.aws_access_key_id = aws_access_key_id self.aws_secret_access_key = aws_secret_access_key def set_sandbox(self, is_sandbox): self.is_sandbox = is_sandbox def get_reviewable_hits(self): if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False reviewable_hits = [ hit for hit in hits if (hit.HITStatus == "Reviewable" or hit.HITStatus == "Reviewing") ] hits_data = [ MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in reviewable_hits ] return (hits_data) def get_all_hits(self): if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False hits_data = [ MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in hits ] return (hits_data) def get_active_hits(self): if not self.connect_to_turk(): return False # hits = self.mtc.search_hits() try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False active_hits = [hit for hit in hits if not (hit.expired)] hits_data = [ MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in active_hits ] return (hits_data) def get_workers(self, assignmentStatus=None): if not self.connect_to_turk(): return False try: hits = self.mtc.search_hits(sort_direction='Descending', page_size=20) hit_ids = [hit.HITId for hit in hits] workers_nested = [ self.mtc.get_assignments(hit_id, status=assignmentStatus, sort_by='SubmitTime', page_size=100) for hit_id in hit_ids ] workers = [val for subl in workers_nested for val in subl] # Flatten nested lists except MTurkRequestError: return (False) worker_data = [{ 'hitId': worker.HITId, 'assignmentId': worker.AssignmentId, 'workerId': worker.WorkerId, 'submit_time': worker.SubmitTime, 'accept_time': worker.AcceptTime, 'status': worker.AssignmentStatus } for worker in workers] return (worker_data) def bonus_worker(self, assignment_id, amount, reason=""): if not self.connect_to_turk(): return False try: bonus = MTurkConnection.get_price_as_price(amount) assignment = self.mtc.get_assignment(assignment_id)[0] workerId = assignment.WorkerId self.mtc.grant_bonus(workerId, assignment_id, bonus, reason) return True except MTurkRequestError as e: print e return False def approve_worker(self, assignment_id): if not self.connect_to_turk(): return (False) try: self.mtc.approve_assignment(assignment_id, feedback=None) return True except MTurkRequestError: return (False) def reject_worker(self, assignment_id): if not self.connect_to_turk(): return False try: self.mtc.reject_assignment(assignment_id, feedback=None) return True except MTurkRequestError: return (False) def unreject_worker(self, assignment_id): if not self.connect_to_turk(): return False try: self.mtc.approve_rejected_assignment(assignment_id) return True except MTurkRequestError: return False def verify_aws_login(self): if (self.aws_access_key_id == 'YourAccessKeyId') or (self.aws_secret_access_key == 'YourSecretAccessKey'): return False else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict( aws_access_key_id=self.aws_access_key_id, aws_secret_access_key=self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**mturkparams) try: self.mtc.get_account_balance() except MTurkRequestError as e: print(e.error_message) return False else: return True def connect_to_turk(self): if not self.validLogin: print 'Sorry, unable to connect to Amazon Mechanical Turk. AWS credentials invalid.' return False if self.is_sandbox: host = 'mechanicalturk.sandbox.amazonaws.com' else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict(aws_access_key_id=self.aws_access_key_id, aws_secret_access_key=self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**mturkparams) return True def configure_hit(self, hit_config): # configure question_url based on the id experimentPortalURL = hit_config['ad_location'] frameheight = 600 mturkQuestion = ExternalQuestion(experimentPortalURL, frameheight) # Qualification: quals = Qualifications() approve_requirement = hit_config['approve_requirement'] quals.add( PercentAssignmentsApprovedRequirement("GreaterThanOrEqualTo", approve_requirement)) if hit_config['us_only']: quals.add(LocaleRequirement("EqualTo", "US")) # Specify all the HIT parameters self.paramdict = dict(hit_type=None, question=mturkQuestion, lifetime=hit_config['lifetime'], max_assignments=hit_config['max_assignments'], title=hit_config['title'], description=hit_config['description'], keywords=hit_config['keywords'], reward=hit_config['reward'], duration=hit_config['duration'], approval_delay=None, questions=None, qualifications=quals) def check_balance(self): if not self.connect_to_turk(): return ('-') return (self.mtc.get_account_balance()[0]) # TODO (if valid AWS credentials haven't been provided then connect_to_turk() will # fail, not error checking here and elsewhere) def create_hit(self, hit_config): try: if not self.connect_to_turk(): return False self.configure_hit(hit_config) myhit = self.mtc.create_hit(**self.paramdict)[0] self.hitid = myhit.HITId except: return False else: return self.hitid # TODO(Jay): Have a wrapper around functions that serializes them. # Default output should not be serialized. def expire_hit(self, hitid): if not self.connect_to_turk(): return False try: self.mtc.expire_hit(hitid) return True except MTurkRequestError: print "Failed to expire HIT. Please check the ID and try again." return False def dispose_hit(self, hitid): if not self.connect_to_turk(): return False try: self.mtc.dispose_hit(hitid) except Exception, e: print 'Failed to dispose of HIT %s. Make sure there are no assignments remaining to be reviewed' % hitid
from boto.mturk.connection import MTurkConnection print "Content-type: text/html" print form = cgi.FieldStorage() HIT = form.getfirst("ID", "empty") SAND = form.getfirst("sandbox", "empty") ACCESS_ID = "AKIAIM5D5I7RUTGYNI7A" SECRET_KEY = "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" # HIT="2JNL8I9NZW6HG96GKYHWCT87ATCVL9" # HOST = 'mechanicalturk.amazonaws.com' # https://mechanicalturk.amazonaws.com/?Service=AWSMechanicalTurkRequester if SAND == 1: print "sandbox" HOST = "mechanicalturk.sandbox.amazonaws.com" else: print "not Sandbox" HOST = "mechanicalturk.amazonaws.com" mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) print mtc.get_account_balance() mtc.extend_hit(hit_id=HIT, assignments_increment=1)
def get_account_balance(): mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) return mtc.get_account_balance()
class MTurkServices: def __init__(self, aws_access_key_id, aws_secret_access_key, is_sandbox): self.update_credentials(aws_access_key_id, aws_secret_access_key) self.set_sandbox(is_sandbox) self.validLogin = self.verify_aws_login() if not self.validLogin: print 'WARNING *****************************' print 'Sorry, AWS Credentials invalid.\nYou will only be able to '\ + 'test experiments locally until you enter\nvalid '\ + 'credentials in the AWS Access section of ~/.psiturkconfig\n' def update_credentials(self, aws_access_key_id, aws_secret_access_key): self.aws_access_key_id = aws_access_key_id self.aws_secret_access_key = aws_secret_access_key def set_sandbox(self, is_sandbox): self.is_sandbox = is_sandbox def get_reviewable_hits(self): if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False reviewable_hits = [hit for hit in hits if (hit.HITStatus == "Reviewable" or hit.HITStatus == "Reviewing")] hits_data = [MTurkHIT({'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in reviewable_hits] return(hits_data) def get_all_hits(self): if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False hits_data = [MTurkHIT({'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in hits] return(hits_data) def get_active_hits(self): if not self.connect_to_turk(): return False # hits = self.mtc.search_hits() try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False active_hits = [hit for hit in hits if not(hit.expired)] hits_data = [MTurkHIT({'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in active_hits] return(hits_data) def get_workers(self, assignmentStatus = None): if not self.connect_to_turk(): return False try: hits = self.mtc.search_hits(sort_direction='Descending', page_size=20) hit_ids = [hit.HITId for hit in hits] workers_nested = [self.mtc.get_assignments( hit_id, status=assignmentStatus, sort_by='SubmitTime', page_size=100 ) for hit_id in hit_ids] workers = [val for subl in workers_nested for val in subl] # Flatten nested lists except MTurkRequestError: return(False) worker_data = [{'hitId': worker.HITId, 'assignmentId': worker.AssignmentId, 'workerId': worker.WorkerId, 'submit_time': worker.SubmitTime, 'accept_time': worker.AcceptTime, 'status': worker.AssignmentStatus } for worker in workers] return(worker_data) def bonus_worker(self, assignment_id, amount, reason=""): if not self.connect_to_turk(): return False try: bonus = MTurkConnection.get_price_as_price(amount) assignment = self.mtc.get_assignment(assignment_id)[0] workerId = assignment.WorkerId self.mtc.grant_bonus(workerId, assignment_id, bonus, reason) return True except MTurkRequestError as e: print e return False def approve_worker(self, assignment_id): if not self.connect_to_turk(): return(False) try: self.mtc.approve_assignment(assignment_id, feedback=None) return True except MTurkRequestError: return(False) def reject_worker(self, assignment_id): if not self.connect_to_turk(): return False try: self.mtc.reject_assignment(assignment_id, feedback=None) return True except MTurkRequestError: return(False) def unreject_worker(self, assignment_id): if not self.connect_to_turk(): return False try: self.mtc.approve_rejected_assignment(assignment_id) return True except MTurkRequestError: return False def verify_aws_login(self): if (self.aws_access_key_id == 'YourAccessKeyId') or (self.aws_secret_access_key == 'YourSecretAccessKey'): return False else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict( aws_access_key_id=self.aws_access_key_id, aws_secret_access_key=self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**mturkparams) try: self.mtc.get_account_balance() except MTurkRequestError as e: print(e.error_message) return False else: return True def connect_to_turk(self): if not self.validLogin: print 'Sorry, unable to connect to Amazon Mechanical Turk. AWS credentials invalid.' return False if self.is_sandbox: host = 'mechanicalturk.sandbox.amazonaws.com' else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict( aws_access_key_id = self.aws_access_key_id, aws_secret_access_key = self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**mturkparams) return True def configure_hit(self, hit_config): # configure question_url based on the id experimentPortalURL = hit_config['ad_location'] frameheight = 600 mturkQuestion = ExternalQuestion(experimentPortalURL, frameheight) # Qualification: quals = Qualifications() approve_requirement = hit_config['approve_requirement'] quals.add( PercentAssignmentsApprovedRequirement("GreaterThanOrEqualTo", approve_requirement)) if hit_config['us_only']: quals.add(LocaleRequirement("EqualTo", "US")) # Specify all the HIT parameters self.paramdict = dict( hit_type = None, question = mturkQuestion, lifetime = hit_config['lifetime'], max_assignments = hit_config['max_assignments'], title = hit_config['title'], description = hit_config['description'], keywords = hit_config['keywords'], reward = hit_config['reward'], duration = hit_config['duration'], approval_delay = None, questions = None, qualifications = quals ) def check_balance(self): if not self.connect_to_turk(): return('-') return(self.mtc.get_account_balance()[0]) # TODO (if valid AWS credentials haven't been provided then connect_to_turk() will # fail, not error checking here and elsewhere) def create_hit(self, hit_config): try: if not self.connect_to_turk(): return False self.configure_hit(hit_config) myhit = self.mtc.create_hit(**self.paramdict)[0] self.hitid = myhit.HITId except: return False else: return self.hitid # TODO(Jay): Have a wrapper around functions that serializes them. # Default output should not be serialized. def expire_hit(self, hitid): if not self.connect_to_turk(): return False try: self.mtc.expire_hit(hitid) return True except MTurkRequestError: print "Failed to expire HIT. Please check the ID and try again." return False def dispose_hit(self, hitid): if not self.connect_to_turk(): return False try: self.mtc.dispose_hit(hitid) except Exception, e: print 'Failed to dispose of HIT %s. Make sure there are no assignments remaining to be reviewed' % hitid
# https://mechanicalturk.amazonaws.com/?Service=AWSMechanicalTurkRequester if SAND==1: print "sandbox" HOST = 'mechanicalturk.sandbox.amazonaws.com' else: print "not Sandbox" HOST = 'mechanicalturk.amazonaws.com' conn = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) print conn.get_account_balance() def accept_and_pay(worker_id, assign_id, bonus_price, reason): """pays for assignment; returns False if something went wrong, else True""" try: #result = conn.approve_assignment(assign_id) #TODO: make sure to avoid the possibility of paying the same bonus twice if bonus_price > 0: conn.grant_bonus(worker_id, assign_id, Price(amount=bonus_price), reason) except MTurkRequestError: #TODO: less embarrasing error handling print "looks like this one was already paid for. or, any other error" return False# no bonus if already paid for return True
class TranscriptionPipelineHandler(): def __init__(self): aws_id = os.environ['AWS_ACCESS_KEY_ID'] aws_k = os.environ['AWS_ACCESS_KEY'] self.conn = MTurkConnection(aws_access_key_id=aws_id,\ aws_secret_access_key=aws_k,\ host=HOST) self.ah = AssignmentHandler(self.conn) self.th = TurkerHandler(self.conn) self.hh = HitHandler(self.conn, TEMPLATE_DIR) self.mh = MongoTranscriptionHandler() self.wh = WavHandler() self.ph = PromptHandler() self.filter = Filter(self.mh) self.balance = self.conn.get_account_balance()[0].amount self.logger = logging.getLogger( "transcription_engine.transcription_pipeline_handler") def audio_clip_referenced_to_hit(self, priority=1, max_queue_size=10): for audio_clip in self.mh.get_artifacts_by_state( "audio_clips", "Referenced"): audio_clip_id = audio_clip["_id"] self.mh.queue_clip(audio_clip_id, priority, max_queue_size) response = self.audio_clip_queue_to_hit() def audio_clip_queued_to_hit(self, priority=1, max_queue_size=10): for audio_clip in self.mh.get_artifacts("audio_clips", {"state": "Queued"}): audio_clip_id = audio_clip["_id"] response = self.audio_clip_queue_to_hit() #=================================================================== # elif state == "Hit": # print("In hit: %s"%audio_clip_url) #=================================================================== def audio_clip_queue_to_hit(self, cost_sensitive=True): """Take queued audio clips from the audio clip queue put them in a hit and create the hit. If successful, update the audio clip state.""" clip_queue = self.mh.get_audio_clip_queue() clip_pairs = self.mh.get_audio_clip_pairs(clip_queue) if clip_pairs: hit_title = "Audio Transcription" question_title = "List and Transcribe" description = "Transcribe the audio clip by typing the words the person says in order." keywords = "audio, transcription, audio transcription" if cost_sensitive: reward_per_clip = 0.02 max_assignments = 3 estimated_cost = self.hh.estimate_html_HIT_cost( clip_pairs, reward_per_clip, max_assignments) clips_in_hits = self.mh.clips_already_in_hit(clip_pairs) if clips_in_hits: #If one or more clips are already in a HIT, remove it from the queue self.mh.remove_audio_clips_from_queue(clips_in_hits) elif self.balance - estimated_cost >= 250: #if we have enough money, create the HIT response = self.hh.make_html_transcription_HIT( clip_pairs, hit_title, question_title, description, keywords) self.balance = self.balance - estimated_cost if type(response) == ResultSet and len( response) == 1 and response[0].IsValid: response = response[0] self.mh.remove_audio_clips_from_queue(clip_queue) audio_clip_ids = [ w["audio_clip_id"] for w in clip_queue ] hit_id = response.HITId hit_type_id = response.HITTypeId self.mh.create_transcription_hit_artifact( hit_id, hit_type_id, clip_queue, "New") self.logger.info("Successfully created HIT: %s" % hit_id) return self.mh.update_audio_clips_state( audio_clip_ids, "Hit") else: pass return False def load_assignments_hit_to_submitted(self): """Check all assignments for audio clip IDs. Update the audio clips. This is a non-destructive load of the assignments from MTurk""" hits = self.conn.get_all_hits() for hit in hits: transcription_dicts = [{}] hit_id = hit.HITId assignments = self.conn.get_assignments(hit_id) have_all_assignments = True assignment_ids = [] for assignment in assignments: assignment_ids.append(assignment.AssignmentId) if self.mh.get_artifact("assignments", {"_id": assignment.AssignmentId}): #We create assignments here, so if we already have it, skip continue else: have_all_assignments = False transcription_ids = [] transcription_dicts = self.ah.get_assignment_submitted_transcriptions( assignment) if transcription_dicts and len(transcription_dicts) == 10: pass for transcription in transcription_dicts: if not self.mh.get_artifact_by_id( "audio_clips", transcription["audio_clip_id"]): self.logger.info("Assignment(%s) with unknown audio clip(%s) skipped"%\ (assignment.AssignmentId,transcription["audio_clip_id"])) break self.mh.update_transcription_state(transcription, "Submitted") self.mh.update_audio_clips_state( [transcription["audio_clip_id"]], "Submitted") transcription_ids.append( self.mh.get_artifact( "transcriptions", { "audio_clip_id": transcription["audio_clip_id"], "assignment_id": transcription["assignment_id"] }, "_id")) else: self.mh.create_assignment_artifact(assignment, transcription_ids, "Submitted") if assignments and not have_all_assignments: self.mh.update_transcription_hit_state(hit_id, "Submitted") print("Transcriptions HIT(%s) submitted assignments: %s " % (hit_id, assignment_ids)) def assignment_submitted_approved(self): """For all submitted assignments, if an answered question has a reference transcription, check the WER. If all the answered questions with reference transcriptions have an acceptable WER, approve the assignment and update the audio clips and transcriptions.""" assignments = self.mh.get_artifacts_by_state("assignments", "Submitted") rejected_feedback = "I'm sorry but your work in assignment(%s) was rejected because" +\ " one or more of your transcriptions " +\ " had a word error rate above the maximum acceptable"+\ " word error rate of %s. Omitted words and words that "+\ " differed by more than %s "+\ " characters were counted as an error." accepted_feedback = "Your average word error rate on assignment(%s) was %s."+\ " Assignment accepted! Thanks for your hard work." for assignment in assignments: assignment_id = assignment["_id"] transcription_ids = assignment["transcriptions"] transcriptions = self.mh.get_artifacts("transcriptions", "_id", transcription_ids) worker_id = assignment["worker_id"] worker_id = self.mh.create_worker_artifact(worker_id) approved, average_wer = self.filter.approve_assignment( transcriptions) if approved: try: self.conn.approve_assignment( assignment_id, accepted_feedback % (assignment_id, average_wer)) except MTurkRequestError as e: print(e) else: self.mh.update_assignment_state(assignment, "Approved") for transcription in transcriptions: #Approve transcriptions without references in the same assignment reference_id = self.mh.get_artifact_by_id( "audio_clips", transcription["audio_clip_id"], "reference_transcription_id") if not reference_id: self.mh.update_transcription_state( transcription, "Approved") print("Approved transcription ids: %s" % transcription_ids) else: #Don't deny for now feedback = rejected_feedback % (assignment_id, self.filter.WER_THRESHOLD, self.filter.CER_THRESHOLD) self.logger.info(feedback) self.conn.reject_assignment(assignment_id, feedback) self.mh.update_assignment_state(assignment, "Denied") #print("Assignments not aproved %s "%denied) #Update the worker if approved: self.mh.add_assignment_to_worker(worker_id, (assignment_id, average_wer)) def _load_rm_audio_source_file_to_clipped( self, file_dir, prompt_file_uri, base_clip_dir, sample_rate=16000, http_base_url="http://www.cis.upenn.edu/~tturpen/wavs/", init_clip_count=200): """For an audio directory, see which files are new and not an audio source already """ prompt_dict = self.ph.get_prompts(prompt_file_uri) count = 0 for root, dirs, files in os.walk(file_dir): for f in files: if count == init_clip_count: return system_uri = os.path.join(root, f) out_uri = system_uri.strip(".sph") + ".wav" out_uri = os.path.basename(out_uri) out_uri = os.path.join(root, (out_uri)) spkr_id = str(os.path.relpath(root, file_dir)) #sph to wav if not f.endswith(".wav") and not os.path.exists(out_uri): try: self.wh.sph_to_wav(system_uri, out_uri=out_uri) except WavHandlerException as e: self.logger.error("Unable to create wav from sph: " + str(e)) if os.path.exists(out_uri) and out_uri.endswith(".wav"): #create audio source artifact count += 1 wav_filename = os.path.basename(out_uri) prompt_id = os.path.basename(out_uri).strip(".wav").upper() encoding = ".wav" sample_rate = 16000 disk_space = os.stat(out_uri).st_size length_seconds = self.wh.get_audio_length(out_uri) if prompt_id in prompt_dict: transcription_prompt = prompt_dict[prompt_id] else: #No prompt found raise PromptNotFound source_id = self.mh.create_audio_source_artifact( out_uri, disk_space, length_seconds, sample_rate, spkr_id, encoding) #create audio clip artifact audio_clip_uri = os.path.join(base_clip_dir, spkr_id, wav_filename) clip_dir = os.path.dirname(audio_clip_uri) if not os.path.exists(clip_dir): os.makedirs(clip_dir) if not os.path.exists(audio_clip_uri): copyfile(out_uri, audio_clip_uri) #http_url http_url = os.path.join(http_base_url, spkr_id, wav_filename) clip_id = self.mh.create_audio_clip_artifact( source_id, 0, -1, audio_clip_uri, http_url, length_seconds, disk_space) #Update the audio source, updates state too self.mh.update_audio_source_audio_clip(source_id, clip_id) #Create the reference transcription artifact transcription_id = self.mh.create_reference_transcription_artifact( clip_id, transcription_prompt, "Gold") #Completes audio clip to Referenced self.mh.update_audio_clip_reference_transcription( clip_id, transcription_id) def all_workers_liveness(self): workers = self.mh.get_all_workers() for worker in workers: worker_id = worker["_id"] approved, denied = self.mh.get_worker_assignments(worker) print("Worker(%s) assignments, approved(%s) denied(%s)" % (worker["_id"], approved, denied)) selection = input( "1. Show denied transcriptions and references.\n" + "2. Show accepted transcriptions and references.\n" + "3. Show both denied and accepted transcriptions.") if selection == 1 or selection == 3: print("Approved transcriptions") for assignment_id in approved: transcription_pairs = self.mh.get_transcription_pairs( assignment_id) for pair in transcription_pairs: print("Reference:\n\t%s\nHypothesis:\n\t%s\n" % (pair[0], pair[1])) if selection == 2 or selection == 3: print("Denied transcriptions") for assignment_id in denied: transcription_pairs = self.mh.get_transcription_pairs( assignment_id) for pair in transcription_pairs: print("Reference:\n\t%s\nHypothesis:\n\t%s\n" % (pair[0], pair[1])) def stats(self): workers = self.mh.get_all_workers() all_wer_per_approved_assignment = 0.0 total_accepted = 0.0 for worker in workers: worker_wer = 0.0 worker_id = worker["_id"] approved, denied = self.mh.get_worker_assignments_wer(worker) for w in approved: all_wer_per_approved_assignment += float(w[1]) worker_wer += float(w[1]) total_accepted += 1 if approved: worker_average_wer = worker_wer / len(approved) print("%s,%s" % (len(approved), worker_average_wer)) #print("Worker(%s) approved assignments(%s)\n denied assignments(%s)"%(worker_id,approved,denied)) av = all_wer_per_approved_assignment / total_accepted print("Average WER per assignment(%s)" % (av)) def get_assignment_stats(self): self.effective_hourly_wage_for_approved_assignments(.20) def effective_hourly_wage_for_approved_assignments(self, reward_per_assignment): """Calculate the effective hourly wage for Approved Assignments""" approved_assignments = self.mh.get_artifacts_by_state( "assignments", "Approved") total = datetime.timedelta(0) count = 0 for assignment in approved_assignments: if "SubmitTime" in assignment: accepted = datetime.datetime.strptime(assignment["AcceptTime"], "%Y-%m-%dT%H:%M:%SZ") submitted = datetime.datetime.strptime( assignment["SubmitTime"], "%Y-%m-%dT%H:%M:%SZ") else: pass total += submitted - accepted count += 1 seconds_per_assignment = total.total_seconds() / count effective_hourly_wage = 60.0 * 60.0 / seconds_per_assignment * reward_per_assignment print("Effective completion time(%s) *reward(%s) = %s" % (seconds_per_assignment, reward_per_assignment, effective_hourly_wage)) def allhits_liveness(self): #allassignments = self.conn.get_assignments(hit_id) #first = self.ah.get_submitted_transcriptions(hit_id,str(clipid)) hits = self.conn.get_all_hits() for hit in hits: hit_id = hit.HITId print("HIT ID: %s" % hit_id) assignments = self.conn.get_assignments(hit_id) if len(assignments) == 0: if raw_input("Remove hit with no submitted assignments?(y/n)" ) == "y": try: self.conn.disable_hit(hit_id) clips = self.mh.get_artifact("transcription_hits", {"_id": hit_id}, "clips") self.mh.remove_transcription_hit(hit_id) self.mh.update_audio_clips_state(clips, "Referenced") except MTurkRequestError as e: raise e else: if raw_input("Remove hit with %s submitted assignments?(y/n)" % len(assignments)) == "y": try: self.conn.disable_hit(hit_id) except MTurkRequestError as e: raise e def run(self): audio_file_dir = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/ind_trn" #audio_file_dir = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/dep_trn" prompt_file_uri = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/doc/al_sents.snr" base_clip_dir = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/clips" selection = 0 init_clip_count = 10000 while selection != "11": selection = raw_input( """Audio Source file to Audio Clip Approved Pipeline:\n 1: AudioSource-FileToClipped: Initialize Resource Management audio source files to %d queueable(Referenced) clips 2: AudioClip-ReferencedToHit: Queue all referenced audio clips and create a HIT if the queue is full. 3: AudioClip-HitToSubmitted: Check all submitted assignments for Transcriptions. 4: AudioClip-SubmittedToApproved: Check all submitted clips against their reference. 5: Review Current Hits 6: Worker liveness 7: Account balance 8: Worker stats 9: Recalculate worker WER 10: Assignment Stats 11: Exit """ % init_clip_count) #selection = "5" if selection == "1": self._load_rm_audio_source_file_to_clipped( audio_file_dir, prompt_file_uri, base_clip_dir, init_clip_count=init_clip_count) elif selection == "2": self.audio_clip_referenced_to_hit() elif selection == "3": self.load_assignments_hit_to_submitted() elif selection == "4": self.assignment_submitted_approved() elif selection == "5": self.allhits_liveness() elif selection == "6": self.all_workers_liveness() elif selection == "7": print("Account balance: %s" % self.balance) elif selection == "8": self.stats() elif selection == "9": self.recalculate_worker_assignment_wer() elif selection == "10": self.get_assignment_stats() # def get_time_submitted_for_assignments(self): # assignments = self.mh.get_all_artifacts("assignments") # for assignment in assignments: # assignment_id = assignment["_id"] # a_assignment = self.conn.get_assignment(assignment_id)[0] # self.mh.update_artifact_by_id("assignments", assignment_id, "SubmitTime", a_assignment.SubmitTime) # def recalculate_worker_assignment_wer(self): # """For all submitted assignments, # if an answered question has a reference transcription, # check the WER. # If all the answered questions with reference transcriptions # have an acceptable WER, approve the assignment and update # the audio clips and transcriptions.""" # assignments = self.mh.get_artifacts("assignments",{"state":"Approved"}) # for assignment in assignments: # assignment_id = assignment["_id"] # denied = [] # #If no transcriptions have references then we automatically approve the HIT # approved = True # transcription_ids = assignment["transcriptions"] # transcriptions = self.mh.get_transcriptions("_id",transcription_ids) # worker_id = assignment["worker_id"] # worker_id = self.mh.create_worker_artifact(worker_id) # # max_rej_wer = (0.0,0.0) # total_wer = 0.0 # for transcription in transcriptions: # #Normalize the transcription # #self.mh.normalize_transcription # reference_id = self.mh.get_audio_clip_by_id(transcription["audio_clip_id"],"reference_transcription_id") # if reference_id: # reference_transcription = self.mh.get_reference_transcription({"_id": reference_id}, # "transcription") # new_transcription = transcription["transcription"].split(" ") # if reference_transcription: # transcription_wer = cer_wer(reference_transcription,new_transcription) # total_wer += transcription_wer # if transcription_wer < WER_THRESHOLD: # self.logger.info("WER for transcription(%s) %d"%(transcription["transcription"],transcription_wer)) # else: # max_rej_wer = (transcription_wer,WER_THRESHOLD) # denied.append((reference_transcription,new_transcription)) # approved = False # average_wer = total_wer/len(transcriptions) # #Update the worker # self.mh.add_assignment_to_worker(worker_id,(assignment_id,average_wer))
class ElicitationPipelineHandler(object): def __init__(self): aws_id = os.environ['AWS_ACCESS_KEY_ID'] aws_k = os.environ['AWS_ACCESS_KEY'] try: self.conn = MTurkConnection(aws_access_key_id=aws_id,\ aws_secret_access_key=aws_k,\ host=HOST) except Exception as e: print(e) self.ah = AssignmentHandler(self.conn) self.th = TurkerHandler(self.conn) self.hh = HitHandler(self.conn, TEMPLATE_DIR) self.mh = MongoElicitationHandler() self.ph = PromptHandler() self.filter = Filter(self.mh) self.balance = self.conn.get_account_balance()[0].amount self.batch_cost = 1 if self.balance > self.batch_cost: self.balance = self.batch_cost else: raise IOError self.logger = logging.getLogger( "transcription_engine.elicitation_pipeline_handler") def load_PromptSource_RawToList(self, prompt_file_uri): """Create the prompt artifacts from the source.""" prompt_dict = self.ph.get_prompts(prompt_file_uri) disk_space = os.stat(prompt_file_uri).st_size source_id = self.mh.create_prompt_source_artifact( prompt_file_uri, disk_space, len(prompt_dict)) normalizer = Normalize() for key in prompt_dict: prompt, line_number = prompt_dict[key] normalized_prompt = normalizer.rm_prompt_normalization(prompt) self.mh.create_prompt_artifact(source_id, prompt, normalized_prompt, line_number, key, len(prompt)) def load_assignment_hit_to_submitted(self): """Check all assignments for audio clip IDs. Update the audio clips. This is a non-destructive load of the assignments from MTurk""" hits = self.conn.get_all_hits() for hit in hits: transcription_dicts = [{}] hit_id = hit.HITId if self.mh.get_artifact("elicitation_hits", {"_id": hit_id}): assignments = self.conn.get_assignments(hit_id) have_all_assignments = True assignment_ids = [] for assignment in assignments: assignment_id = assignment.AssignmentId assignment_ids.append(assignment_id) if self.mh.get_artifact("elicitation_assignments", {"_id": assignment.AssignmentId}): #We create assignments here, so if we already have it, skip continue #pass else: have_all_assignments = False recording_ids = [] prompt_id_tag = "prompt_id" recording_url_tag = "recording_url" worker_id_tag = "worker_id" recording_dict = self.ah.get_assignment_submitted_text_dict( assignment, prompt_id_tag, recording_url_tag) worker_oid = self.mh.create_worker_artifact( assignment.WorkerId) zipcode = None for recording in recording_dict: if recording[prompt_id_tag] == "zipcode": zipcode = recording[recording_url_tag] continue if not self.mh.get_artifact_by_id( "prompts", recording[prompt_id_tag]): self.logger.info("Assignment(%s) with unknown %s(%s) skipped"%\ (assignment_id,prompt_id_tag,recording[prompt_id_tag])) break recording_id = self.mh.create_recording_source_artifact( recording[prompt_id_tag], recording[recording_url_tag], recording[worker_id_tag]) if not recording_id: self.mh.create_assignment_artifact(assignment, recording_ids, zipcode=zipcode, incomplete=True) break self.mh.add_item_to_artifact_set( "prompts", recording[prompt_id_tag], "recording_sources", recording_id) recording_ids.append(recording_id) else: self.mh.create_assignment_artifact(assignment, recording_ids, zipcode=zipcode) self.mh.add_item_to_artifact_set( "elicitation_hits", hit_id, "submitted_assignments", assignment_id) self.mh.add_item_to_artifact_set( "workers", worker_oid, "submitted_assignments", assignment_id) print("Elicitation HIT(%s) submitted assignments: %s " % (hit_id, assignment_ids)) def approve_assignment_submitted_to_approved(self): """Approve all submitted assignments""" hits = self.conn.get_all_hits() for hit in hits: transcription_dicts = [{}] hit_id = hit.HITId if self.mh.get_artifact("elicitation_hits", {"_id": hit_id}): assignments = self.conn.get_assignments(hit_id) have_all_assignments = True assignment_ids = [] for assignment in assignments: assignment_id = assignment.AssignmentId assignment_ids.append(assignment_id) if self.mh.get_artifact("elicitation_assignments", { "_id": assignment_id, "state": "Submitted" }): #WARNING: this Approves every assignment self.conn.approve_assignment( assignment_id, "Thank you for completing this assignment!") self.mh.update_artifact_by_id( "elicitation_assignments", assignment_id, "approval_time", datetime.datetime.now()) def approve_assignment_by_worker(self): """Approve all submitted assignments""" approval_comment = "Thank you for your recordings, good work, assignment approved!" denial_comment = "I'm sorry but your work was denied because %s" hits = self.conn.get_all_hits() for hit in hits: transcription_dicts = [{}] hit_id = hit.HITId if self.mh.get_artifact("elicitation_hits", {"_id": hit_id}): assignments = self.conn.get_assignments(hit_id) have_all_assignments = True assignment_ids = [] for assignment in assignments: assignment_id = assignment.AssignmentId assignment_ids.append(assignment_id) if self.mh.get_artifact("elicitation_assignments", { "_id": assignment_id, "state": "Submitted" }): #WARNING: this Approves every assignment assignment_artifact = self.mh.get_artifact( "elicitation_assignments", {"_id": assignment_id}) recording_ids = assignment_artifact["recordings"] worker = self.mh.get_artifact( "workers", {"eid": assignment_artifact["worker_id"]}) if worker["state"] == "Approved": #If the worker is approved, approve the assignment automatically self.conn.approve_assignment( assignment_id, approval_comment) self.mh.update_artifact_by_id( "elicitation_assignments", assignment_id, "approval_time", datetime.datetime.now()) continue elif worker["state"] == "Rejected": self.conn.reject_assignment( assignment_id, worker["rejection_reason"]) self.mh.update_artifact_by_id( "elicitation_assignments", assignment_id, "approval_time", datetime.datetime.now()) continue recording_uris = [] for recording_id in recording_ids: uri = self.mh.get_artifact_by_id( "recording_sources", recording_id, "recording_uri") recording_uris.append(uri) command = ["gnome-mplayer"] + recording_uris if len(recording_uris) > 0 and recording_uris[ 0].endswith(" .wav") or recording_uris[ 0].endswith(".com.wav"): continue print("Calling: %s" % command) call(command) approve_assignment = raw_input( "Approve assignment(y/n/s)?") if approve_assignment == "s": #skip continue elif approve_assignment == "y": #accept the assignment self.conn.approve_assignment( assignment_id, approval_comment) self.mh.update_artifact_by_id( "elicitation_assignments", assignment_id, "approval_time", datetime.datetime.now()) approve_worker = raw_input("Approve worker(y/n)?") if approve_worker == "y": #approve the worker and all future assignments self.mh.update_artifact_by_id( "workers", worker["_id"], "approval_time", datetime.datetime.now()) elif approve_assignment == "n": #Reject the assignment reject_worker = raw_input( "Reject this worker's future work?") if reject_worker == "y": #Reject the worker reason = raw_input( "Reason for rejecting this worker's future work:" ) self.mh.update_artifact_by_id( "workers", worker["_id"], "rejection_reason", reason) self.conn.reject_assignment( assignment_id, denial_comment % reason + ".") else: reason = raw_input( "Why reject the assignment?") self.conn.reject_assignment( assignment_id, denial_comment % reason + ".") def get_assignment_stats(self): effective_hourly_wage = self.effective_hourly_wage_for_approved_assignments( .20) def effective_hourly_wage_for_approved_assignments(self, reward_per_assignment): """Calculate the effective hourly wage for Approved Assignments""" approved_assignments = self.mh.get_artifacts_by_state( "elicitation_assignments", "Approved") total = datetime.timedelta(0) count = 0 for assignment in approved_assignments: accepted = datetime.datetime.strptime(assignment["AcceptTime"], "%Y-%m-%dT%H:%M:%SZ") submitted = datetime.datetime.strptime(assignment["SubmitTime"], "%Y-%m-%dT%H:%M:%SZ") total += submitted - accepted count += 1 #self.mh.update_artifact_by_id("elicitation_assignments", assignment["_id"], "SubmitTime", completion_time) seconds_per_assignment = total.total_seconds() / count effective_hourly_wage = 60.0 * 60.0 / seconds_per_assignment * reward_per_assignment print("Effective completion time(%s) *reward(%s) = %s" % (seconds_per_assignment, reward_per_assignment, effective_hourly_wage)) def enqueue_prompts_and_generate_hits(self): prompts = self.mh.get_artifacts_by_state("prompts", "New") for prompt in prompts: self.mh.enqueue_prompt(prompt["_id"], 1, 5) prompt_queue = self.mh.get_prompt_queue() prompt_pairs = self.mh.get_prompt_pairs(prompt_queue) if prompt_pairs: hit_title = "Audio Elicitation" question_title = "Speak and Record your Voice" hit_description = "Speak the prompt and record your voice." keywords = "audio, elicitation, speech, recording" if cost_sensitive: reward_per_clip = 0.04 max_assignments = 2 estimated_cost = self.hh.estimate_html_HIT_cost(prompt_pairs,reward_per_clip=reward_per_clip,\ max_assignments=max_assignments) prompts_in_hits = self.mh.prompts_already_in_hit( prompt_pairs) if prompts_in_hits: #If one or more clips are already in a HIT, remove it from the queue self.mh.remove_artifact_from_queue(prompts_in_hits) elif self.balance - estimated_cost >= 0: #if we have enough money, create the HIT response = self.hh.make_html_elicitation_HIT( prompt_pairs, hit_title, question_title, keywords, hit_description, max_assignments=max_assignments, reward_per_clip=reward_per_clip) # response = self.hh.make_question_form_elicitation_HIT(prompt_pairs,hit_title, # question_title, keywords) self.balance = self.balance - estimated_cost if type(response) == ResultSet and len( response) == 1 and response[0].IsValid: response = response[0] self.mh.remove_artifacts_from_queue( "prompt_queue", prompt_queue) prompt_ids = [w["prompt_id"] for w in prompt_queue] hit_id = response.HITId hit_type_id = response.HITTypeId self.mh.create_elicitation_hit_artifact( hit_id, hit_type_id, prompt_ids) self.mh.update_artifacts_by_id( "prompts", prompt_ids, "hit_id", hit_id) self.logger.info("Successfully created HIT: %s" % hit_id) else: return True print("Amount left in batch: %s out of %s" % (self.balance, self.batch_cost)) def allhits_liveness(self): #allassignments = self.conn.get_assignments(hit_id) #first = self.ah.get_submitted_transcriptions(hit_id,str(clipid)) hits = self.conn.get_all_hits() selection = raw_input("Remove all hits with no assignments?") if selection == "y": for hit in hits: hit_id = hit.HITId assignments = self.conn.get_assignments(hit_id) if len(assignments) == 0: try: self.conn.disable_hit(hit_id) prompts = self.mh.get_artifact("elicitation_hits", {"_id": hit_id}, "prompts") self.mh.remove_elicitation_hit(hit_id) if prompts: self.mh.update_artifacts_state("prompts", prompts) else: pass except MTurkRequestError as e: raise e return True for hit in hits: hit_id = hit.HITId print("HIT ID: %s" % hit_id) assignments = self.conn.get_assignments(hit_id) if len(assignments) == 0: if raw_input("Remove hit with no submitted assignments?(y/n)" ) == "y": try: self.conn.disable_hit(hit_id) prompts = self.mh.get_artifact("elicitation_hits", {"_id": hit_id}, "prompts") self.mh.remove_elicitation_hit(hit_id) if prompts: self.mh.update_artifacts_state("prompts", prompts) else: pass except MTurkRequestError as e: raise e else: if raw_input("Remove hit with %s submitted assignments?(y/n)" % len(assignments)) == "y": try: self.conn.disable_hit(hit_id) except MTurkRequestError as e: raise e def run(self): #audio_file_dir = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/dep_trn" prompt_file_uri = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/doc/al_sents.snr" selection = 0 #self.get_time_submitted_for_assignments() while selection != "8": selection = raw_input( """Prompt Source raw to Elicitations-Approved Pipeline:\n 1: PromptSource-Load_RawToList: Load Resource Management 1 prompt source files to queueable prompts 2: Prompt-ReferencedToHit: Queue all referenced prompts and create a HIT if the queue is full. 3: Prompt-HitToAssignmentSubmitted: Check all submitted assignments for Elicitations and download elicitations. 4: Maintain all assignments and hits. 5: (WARNING, approves all assignments) Approve all submitted assignments. 6: Calculate assignment stats. 7: Hand approve submitted assignments by elicitation and/or by worker. 8: Exit """) if selection == "1": self.load_PromptSource_RawToList(prompt_file_uri) elif selection == "2": self.enqueue_prompts_and_generate_hits() elif selection == "3": self.load_assignment_hit_to_submitted() elif selection == "4": self.allhits_liveness() elif selection == "5": self.approve_assignment_submitted_to_approved() elif selection == "6": self.get_assignment_stats() elif selection == "7": self.approve_assignment_by_worker() else: selection = "8" # prompt_dict = self.ph.get_prompts(prompt_file_uri) # def get_time_submitted_for_assignments(self): # assignments = self.mh.get_all_artifacts("elicitation_assignments") # for assignment in assignments: # assignment_id = assignment["_id"] # a_assignment = self.conn.get_assignment(assignment_id)[0] # self.mh.update_artifact_by_id("elicitation_assignments", assignment_id, "SubmitTime", a_assignment.SubmitTime)
class MTurk(object): def __init__(self, app=None): self.host = 'https://mechanicalturk.sandbox.amazonaws.com' self.secret_key = None self.access_id = None self.app = app if app is not None: self.init_app(app) def init_app(self, app): app.config.setdefault('MTURK_SECRET_KEY', None) app.config.setdefault('MTURK_ACCESS_ID', None) app.config.setdefault('MTURK_SANDBOX', True) self.update_credentials(app.config['MTURK_ACCESS_ID'], app.config['MTURK_SECRET_KEY']) self.is_sandbox = app.config['MTURK_SANDBOX'] self.valid_login = self.verify_aws_login() def update_credentials(self, aws_access_key_id, aws_secret_access_key): self.aws_access_key_id = aws_access_key_id self.aws_secret_access_key = aws_secret_access_key def verify_aws_login(self): if ((self.aws_secret_access_key is None) or (self.aws_access_key_id is None)): logging.warning('No AWS keys found in app configuration') else: host = 'mechanicalturk.amazonaws.com' params = dict(aws_access_key_id=self.aws_access_key_id, aws_secret_access_key=self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**params) try: self.mtc.get_account_balance() except MTurkRequestError as e: return dict(success=False, message=e.error_message) else: return True def connect_to_turk(self): if not self.valid_login: logging.warning( 'Sorry, unable to connect to Amazon Mechanical Turk. Please check your credentials' ) return False if self.is_sandbox: host = 'mechanicalturk.sandbox.amazonaws.com' else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict(aws_access_key_id=self.aws_access_key_id, aws_secret_access_key=self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**mturkparams) return True def get_account_balance(self): if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') try: balance = self.mtc.get_account_balance() except MTurkRequestError as e: return dict(success=False, message=e.error_message) else: return balance def get_reviewable_hits(self): if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') try: hits = self.mtc.get_all_hits() except MTurkRequestError as e: return dict(success=False, message=e.error_message) reviewable_hits = [ hit for hit in hits if hit.HITStatus == "Reviewable" or hit.HITStatus == "Reviewing" ] hits_data = [ MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration }) for hit in reviewable_hits ] return hits_data def get_all_hits(self): """ Get all HITs """ if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() except MTurkRequestError as e: return dict(success=False, message=e.error_message) hits_data = [ MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in hits ] return hits_data def get_active_hits(self): """ Get active HITs """ if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') # hits = self.mtc.search_hits() try: hits = self.mtc.get_all_hits() except MTurkRequestError as e: return dict(success=False, message=e.error_message) active_hits = [hit for hit in hits if not hit.expired] hits_data = [ MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in active_hits ] return hits_data def get_hit(self, hit_id, response_groups=None): if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') try: hit = self.mtc.get_hit(hit_id, response_groups)[0] except MTurkRequestError as e: return False return hit def get_workers(self, assignment_status=None): """ Get workers """ if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') try: hits = self.mtc.search_hits(sort_direction='Descending', page_size=20) except MTurkRequestError as e: return False hit_ids = [hit.HITId for hit in hits] workers_nested = [ self.mtc.get_assignments(hit_id, status=assignment_status, sort_by='SubmitTime', page_size=100) for hit_id in hit_ids ] workers = [val for subl in workers_nested for val in subl] # Flatten nested lists worker_data = [{ 'hitId': worker.HITId, 'assignmentId': worker.AssignmentId, 'workerId': worker.WorkerId, 'submit_time': worker.SubmitTime, 'accept_time': worker.AcceptTime, 'status': worker.AssignmentStatus, 'completion_code': worker.answers[0][0].fields[0] } for worker in workers] return worker_data def bonus_worker(self, assignment_id, amount, reason=""): """ Bonus worker """ if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') try: bonus = MTurkConnection.get_price_as_price(amount) assignment = self.mtc.get_assignment(assignment_id)[0] worker_id = assignment.WorkerId self.mtc.grant_bonus(worker_id, assignment_id, bonus, reason) return True except MTurkRequestError as e: return dict(success=False, message=e.error_message) def approve_worker(self, assignment_id, feedback=None): """ Approve worker """ if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') try: self.mtc.approve_assignment(assignment_id, feedback=feedback) return True except MTurkRequestError as e: return False def reject_worker(self, assignment_id): """ Reject worker """ if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') try: self.mtc.reject_assignment(assignment_id, feedback=None) return True except MTurkRequestError as e: return dict(success=False, message=e.error_message) def unreject_worker(self, assignment_id): """ Unreject worker """ if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') try: self.mtc.approve_rejected_assignment(assignment_id) return True except MTurkRequestError as e: return dict(success=False, message=e.error_message) def assign_qualification(self, qualification_type_id, worker_id, value=1, send_notification=True): if not self.connect_to_turk(): return dict(success=False, message='Could not connect to AWS') try: self.mtc.assign_qualification(qualification_type_id, worker_id, value, send_notification) return True except MTurkRequestError as e: return dict(success=False, message=e.error_message) def revoke_qualification(self, subject_id, qualification_type_id, reason=None): if not self.connect_to_turk(): return False try: self.mtc.revoke_qualification(subject_id, qualification_type_id, reason) return True except MTurkRequestError as e: return dict(success=False, message=e.error_message) def notify_worker(self, worker_id, subject, message_text): if not self.connect_to_turk(): return False try: self.mtc.notify_workers(worker_id, subject, message_text) return True except MTurkRequestError as e: return dict(success=False, message=e.error_message) def list_workers_with_qualification(self, qualification_type_id): if not self.connect_to_turk(): return False try: workers = self.mtc.get_all_qualifications_for_qual_type( qualification_type_id) except MTurkRequestError as e: return dict(success=False, message=e.error_message) workers = [w.SubjectId for w in workers] return workers
class mTurk(): ACCESS_ID ='' SECRET_KEY = '' HOST = '' mtc = None log=None def __init__(self,access_id,secret_key,debug=True): self.ACCESS_ID=access_id self.SECRET_KEY=secret_key if debug: self.HOST='mechanicalturk.sandbox.amazonaws.com' else: raise Exception('sure you want to spend money') self.mtc = MTurkConnection(aws_access_key_id=self.ACCESS_ID, aws_secret_access_key=self.SECRET_KEY, host=self.HOST) def getBalance(self): return self.mtc.get_account_balance() def hasEnoughMoney(self,cost): return cost<self.getBalance() def approveAssignemnt(self,a_id,feedback="Thanks for contribution"): assignment= self.mtc.get_assignment(a_id)[0] log.debug("%s",assignment.__dict__) if assignment.AssignmentStatus!="Approved": ret = self.mtc.approve_assignment(a_id, feedback) log.debug("approve ret: %s",ret) else: log.debug("already approvred") def rejectAssigment(self,a_id,feedback="Sorry, but your work was not satifactory"): ret = self.mtc.reject_assignment(a_id, feedback) log.debug("reject ret: %s",ret) def createExternalQuestion(self,title,description,keywords,url,duration,reward): ex_q=ExternalQuestion(url,1000) res=self.mtc.create_hit(question=ex_q,max_assignments=1,title=title,description=description,keywords=keywords,duration = duration,reward=reward) log.debug("created external question %s",res[0]) return res[0] def getDataFromHit(self, assignmentId, workerId): assignment= self.mtc.get_assignment(assignmentId)[0] log.debug("Answers of the worker %s", assignment.WorkerId) if assignment.WorkerId == workerId: ret = {} for question_form_answer in assignment.answers[0]: if len(question_form_answer.fields)>1: log.debug("answers are >1") ret[question_form_answer.qid]=question_form_answer.fields else: ret[question_form_answer.qid]=question_form_answer.fields[0] return ret # print question_form_answer.qid," - "," ".join(question_form_answer.fields) # # for r in res: # # print "Your hit ID is this %s -> https://workersandbox.mturk.com/mturk/preview?groupId=%s"%(r.HITId,r.HITTypeId)) # # # def printAllHits(self): # # hits=self.mtc.get_all_hits() # # for hit in hits: # # print printAtt(hit,'HITId') # # # def rejectAllHITs(self): # # hits=self.mtc.get_all_hits() # # for hit in hits: # # assignements = self.mtc.get_assignments(hit.HITId) # # for assignment in assignements: # # try: # # self.mtc.reject_assignment(assignment.AssignmentId, "i'm just testing this functionality") # # print "Rejected the assignment %s of HIT %s"%(assignment.AssignmentId,hit.HITId) # # except Exception: # # print "ERROR with the assignment %s of HIT %s"%(assignment.AssignmentId,hit.HITId) # # # ret = self.mtc.disable_hit(hit.HITId, "HITDetail") ## print ret
frameheight = 600 mturkQuestion = ExternalQuestion( experimentPortalURL, 600 ) # Qualification: quals = Qualifications(); quals.add( PercentAssignmentsApprovedRequirement("GreaterThanOrEqualTo", "95") ) quals.add( LocaleRequirement("EqualTo", "US") ) #quals.add( NumberHitsApprovedRequirement("GreaterThanOrEqualTo", "100") ) # Specify all the HIT parameters paramdict = dict( hit_type = None, question = mturkQuestion, lifetime = datetime.timedelta(1), # How long the HIT will be available max_assignments = 50, # Total times it will be assigned, not max per turker title = "Paid volunteers needed for an online experiment in Psychology", description = "Learn to categorize a set of cards over a series of training trials.", keywords = "New York University, psychology experiment, category learning", reward = 0.75, duration = datetime.timedelta(hours=2), approval_delay = None, annotation = None, # Do we need this? Not clear on what it is. questions = None, qualifications = quals ) myhit = mtc.create_hit(**paramdict )[0] hitid = myhit.HITId print mtc.get_account_balance() # Tests the connection
from boto.mturk.connection import MTurkConnection from boto.mturk.question import QuestionContent, Question, QuestionForm, Overview, AnswerSpecification, SelectionAnswer, FormattedContent, FreeTextAnswer from keys import AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY debug = True HOST = 'mechanicalturk.sandbox.amazonaws.com' PIN_IMAGE_URL = 'http://media-cache-ak0.pinimg.com/236x/17/8f/99/178f993435fb2718ab6e22ba29d704e0.jpg' PIN_IMAGE_TITLE = 'Arnhem Clothing \'Song Bird\' Kimono in Mayan Song. Via Soleilblue..' mtc = MTurkConnection(aws_access_key_id=AWS_ACCESS_KEY_ID, aws_secret_access_key=AWS_SECRET_ACCESS_KEY, host=HOST) if debug: print mtc.get_account_balance() title = 'Match these Pictures to Macy\'s Products' description = 'Look at this photo and match it to Macy\'s products' keywords = 'clothing, rating, opinions, easy, quick, macys' #make overview overview = Overview() overview.append_field('Title', 'Find three Macys.com Product Web IDs That Match') overview.append( FormattedContent('<img src="' + PIN_IMAGE_URL + '" alt="Pintrest Image" />' '<br />' + PIN_IMAGE_TITLE)) #make webid1
class TranscriptionPipelineHandler(): def __init__(self): aws_id = os.environ['AWS_ACCESS_KEY_ID'] aws_k = os.environ['AWS_ACCESS_KEY'] self.conn = MTurkConnection(aws_access_key_id=aws_id,\ aws_secret_access_key=aws_k,\ host=HOST) self.ah = AssignmentHandler(self.conn) self.th = TurkerHandler(self.conn) self.hh = HitHandler(self.conn,TEMPLATE_DIR) self.mh = MongoTranscriptionHandler() self.wh = WavHandler() self.ph = PromptHandler() self.filter = Filter(self.mh) self.balance = self.conn.get_account_balance()[0].amount self.logger = logging.getLogger("transcription_engine.transcription_pipeline_handler") def audio_clip_referenced_to_hit(self,priority=1,max_queue_size=10): for audio_clip in self.mh.get_artifacts_by_state("audio_clips","Referenced"): audio_clip_id = audio_clip["_id"] self.mh.queue_clip(audio_clip_id, priority, max_queue_size) response = self.audio_clip_queue_to_hit() def audio_clip_queued_to_hit(self,priority=1,max_queue_size=10): for audio_clip in self.mh.get_artifacts("audio_clips",{"state":"Queued"}): audio_clip_id = audio_clip["_id"] response = self.audio_clip_queue_to_hit() #=================================================================== # elif state == "Hit": # print("In hit: %s"%audio_clip_url) #=================================================================== def audio_clip_queue_to_hit(self,cost_sensitive=True): """Take queued audio clips from the audio clip queue put them in a hit and create the hit. If successful, update the audio clip state.""" clip_queue = self.mh.get_audio_clip_queue() clip_pairs = self.mh.get_audio_clip_pairs(clip_queue) if clip_pairs: hit_title = "Audio Transcription" question_title = "List and Transcribe" description = "Transcribe the audio clip by typing the words the person says in order." keywords = "audio, transcription, audio transcription" if cost_sensitive: reward_per_clip = 0.02 max_assignments = 3 estimated_cost = self.hh.estimate_html_HIT_cost(clip_pairs,reward_per_clip,max_assignments) clips_in_hits = self.mh.clips_already_in_hit(clip_pairs) if clips_in_hits: #If one or more clips are already in a HIT, remove it from the queue self.mh.remove_audio_clips_from_queue(clips_in_hits) elif self.balance - estimated_cost >= 250: #if we have enough money, create the HIT response = self.hh.make_html_transcription_HIT(clip_pairs,hit_title, question_title, description, keywords) self.balance = self.balance - estimated_cost if type(response) == ResultSet and len(response) == 1 and response[0].IsValid: response = response[0] self.mh.remove_audio_clips_from_queue(clip_queue) audio_clip_ids = [w["audio_clip_id"] for w in clip_queue] hit_id = response.HITId hit_type_id = response.HITTypeId self.mh.create_transcription_hit_artifact(hit_id,hit_type_id,clip_queue,"New") self.logger.info("Successfully created HIT: %s"%hit_id) return self.mh.update_audio_clips_state(audio_clip_ids,"Hit") else: pass return False def load_assignments_hit_to_submitted(self): """Check all assignments for audio clip IDs. Update the audio clips. This is a non-destructive load of the assignments from MTurk""" hits = self.conn.get_all_hits() for hit in hits: transcription_dicts = [{}] hit_id = hit.HITId assignments = self.conn.get_assignments(hit_id) have_all_assignments = True assignment_ids = [] for assignment in assignments: assignment_ids.append(assignment.AssignmentId) if self.mh.get_artifact("assignments",{"_id":assignment.AssignmentId}): #We create assignments here, so if we already have it, skip continue else: have_all_assignments = False transcription_ids = [] transcription_dicts = self.ah.get_assignment_submitted_transcriptions(assignment) if transcription_dicts and len(transcription_dicts)==10: pass for transcription in transcription_dicts: if not self.mh.get_artifact_by_id("audio_clips",transcription["audio_clip_id"]): self.logger.info("Assignment(%s) with unknown audio clip(%s) skipped"%\ (assignment.AssignmentId,transcription["audio_clip_id"])) break self.mh.update_transcription_state(transcription,"Submitted") self.mh.update_audio_clips_state([transcription["audio_clip_id"]], "Submitted") transcription_ids.append(self.mh.get_artifact("transcriptions",{"audio_clip_id" : transcription["audio_clip_id"], "assignment_id" : transcription["assignment_id"]}, "_id")) else: self.mh.create_assignment_artifact(assignment, transcription_ids, "Submitted") if assignments and not have_all_assignments: self.mh.update_transcription_hit_state(hit_id,"Submitted") print("Transcriptions HIT(%s) submitted assignments: %s "%(hit_id,assignment_ids)) def assignment_submitted_approved(self): """For all submitted assignments, if an answered question has a reference transcription, check the WER. If all the answered questions with reference transcriptions have an acceptable WER, approve the assignment and update the audio clips and transcriptions.""" assignments = self.mh.get_artifacts_by_state("assignments", "Submitted") rejected_feedback = "I'm sorry but your work in assignment(%s) was rejected because" +\ " one or more of your transcriptions " +\ " had a word error rate above the maximum acceptable"+\ " word error rate of %s. Omitted words and words that "+\ " differed by more than %s "+\ " characters were counted as an error." accepted_feedback = "Your average word error rate on assignment(%s) was %s."+\ " Assignment accepted! Thanks for your hard work." for assignment in assignments: assignment_id = assignment["_id"] transcription_ids = assignment["transcriptions"] transcriptions = self.mh.get_artifacts("transcriptions","_id",transcription_ids) worker_id = assignment["worker_id"] worker_id = self.mh.create_worker_artifact(worker_id) approved, average_wer = self.filter.approve_assignment(transcriptions) if approved: try: self.conn.approve_assignment(assignment_id, accepted_feedback%(assignment_id,average_wer)) except MTurkRequestError as e: print(e) else: self.mh.update_assignment_state(assignment,"Approved") for transcription in transcriptions: #Approve transcriptions without references in the same assignment reference_id = self.mh.get_artifact_by_id("audio_clips",transcription["audio_clip_id"],"reference_transcription_id") if not reference_id: self.mh.update_transcription_state(transcription,"Approved") print("Approved transcription ids: %s"%transcription_ids) else: #Don't deny for now feedback = rejected_feedback%(assignment_id,self.filter.WER_THRESHOLD,self.filter.CER_THRESHOLD) self.logger.info(feedback) self.conn.reject_assignment(assignment_id,feedback) self.mh.update_assignment_state(assignment,"Denied") #print("Assignments not aproved %s "%denied) #Update the worker if approved: self.mh.add_assignment_to_worker(worker_id,(assignment_id,average_wer)) def _load_rm_audio_source_file_to_clipped(self,file_dir,prompt_file_uri, base_clip_dir,sample_rate=16000, http_base_url = "http://www.cis.upenn.edu/~tturpen/wavs/", init_clip_count = 200): """For an audio directory, see which files are new and not an audio source already """ prompt_dict = self.ph.get_prompts(prompt_file_uri) count = 0 for root, dirs, files in os.walk(file_dir): for f in files: if count == init_clip_count: return system_uri = os.path.join(root,f) out_uri = system_uri.strip(".sph") + ".wav" out_uri = os.path.basename(out_uri) out_uri = os.path.join(root,(out_uri)) spkr_id = str(os.path.relpath(root,file_dir)) #sph to wav if not f.endswith(".wav") and not os.path.exists(out_uri): try: self.wh.sph_to_wav(system_uri,out_uri=out_uri) except WavHandlerException as e: self.logger.error("Unable to create wav from sph: "+str(e)) if os.path.exists(out_uri) and out_uri.endswith(".wav"): #create audio source artifact count += 1 wav_filename = os.path.basename(out_uri) prompt_id = os.path.basename(out_uri).strip(".wav").upper() encoding = ".wav" sample_rate = 16000 disk_space = os.stat(out_uri).st_size length_seconds = self.wh.get_audio_length(out_uri) if prompt_id in prompt_dict: transcription_prompt = prompt_dict[prompt_id] else: #No prompt found raise PromptNotFound source_id = self.mh.create_audio_source_artifact(out_uri, disk_space, length_seconds, sample_rate, spkr_id, encoding) #create audio clip artifact audio_clip_uri = os.path.join(base_clip_dir,spkr_id,wav_filename) clip_dir = os.path.dirname(audio_clip_uri) if not os.path.exists(clip_dir): os.makedirs(clip_dir) if not os.path.exists(audio_clip_uri): copyfile(out_uri,audio_clip_uri) #http_url http_url = os.path.join(http_base_url,spkr_id,wav_filename) clip_id = self.mh.create_audio_clip_artifact(source_id, 0, -1, audio_clip_uri, http_url, length_seconds, disk_space) #Update the audio source, updates state too self.mh.update_audio_source_audio_clip(source_id,clip_id) #Create the reference transcription artifact transcription_id = self.mh.create_reference_transcription_artifact(clip_id, transcription_prompt, "Gold") #Completes audio clip to Referenced self.mh.update_audio_clip_reference_transcription(clip_id,transcription_id) def all_workers_liveness(self): workers = self.mh.get_all_workers() for worker in workers: worker_id = worker["_id"] approved, denied = self.mh.get_worker_assignments(worker) print("Worker(%s) assignments, approved(%s) denied(%s)"%(worker["_id"],approved,denied)) selection = input("1. Show denied transcriptions and references.\n"+ "2. Show accepted transcriptions and references.\n"+ "3. Show both denied and accepted transcriptions.") if selection == 1 or selection == 3: print("Approved transcriptions") for assignment_id in approved: transcription_pairs = self.mh.get_transcription_pairs(assignment_id) for pair in transcription_pairs: print ("Reference:\n\t%s\nHypothesis:\n\t%s\n"%(pair[0],pair[1])) if selection == 2 or selection == 3: print("Denied transcriptions") for assignment_id in denied: transcription_pairs = self.mh.get_transcription_pairs(assignment_id) for pair in transcription_pairs: print ("Reference:\n\t%s\nHypothesis:\n\t%s\n"%(pair[0],pair[1])) def stats(self): workers = self.mh.get_all_workers() all_wer_per_approved_assignment = 0.0 total_accepted = 0.0 for worker in workers: worker_wer = 0.0 worker_id = worker["_id"] approved, denied = self.mh.get_worker_assignments_wer(worker) for w in approved: all_wer_per_approved_assignment += float(w[1]) worker_wer += float(w[1]) total_accepted += 1 if approved: worker_average_wer = worker_wer/len(approved) print("%s,%s"%(len(approved),worker_average_wer)) #print("Worker(%s) approved assignments(%s)\n denied assignments(%s)"%(worker_id,approved,denied)) av = all_wer_per_approved_assignment/total_accepted print("Average WER per assignment(%s)"%(av)) def get_assignment_stats(self): self.effective_hourly_wage_for_approved_assignments(.20) def effective_hourly_wage_for_approved_assignments(self,reward_per_assignment): """Calculate the effective hourly wage for Approved Assignments""" approved_assignments = self.mh.get_artifacts_by_state("assignments","Approved") total = datetime.timedelta(0) count = 0 for assignment in approved_assignments: if "SubmitTime" in assignment: accepted = datetime.datetime.strptime(assignment["AcceptTime"],"%Y-%m-%dT%H:%M:%SZ") submitted = datetime.datetime.strptime(assignment["SubmitTime"],"%Y-%m-%dT%H:%M:%SZ") else: pass total += submitted-accepted count += 1 seconds_per_assignment = total.total_seconds()/count effective_hourly_wage = 60.0*60.0/seconds_per_assignment * reward_per_assignment print("Effective completion time(%s) *reward(%s) = %s"%(seconds_per_assignment,reward_per_assignment,effective_hourly_wage)) def allhits_liveness(self): #allassignments = self.conn.get_assignments(hit_id) #first = self.ah.get_submitted_transcriptions(hit_id,str(clipid)) hits = self.conn.get_all_hits() for hit in hits: hit_id = hit.HITId print("HIT ID: %s"%hit_id) assignments = self.conn.get_assignments(hit_id) if len(assignments) == 0: if raw_input("Remove hit with no submitted assignments?(y/n)") == "y": try: self.conn.disable_hit(hit_id) clips = self.mh.get_artifact("transcription_hits",{"_id": hit_id},"clips") self.mh.remove_transcription_hit(hit_id) self.mh.update_audio_clips_state(clips, "Referenced") except MTurkRequestError as e: raise e else: if raw_input("Remove hit with %s submitted assignments?(y/n)"%len(assignments)) == "y": try: self.conn.disable_hit(hit_id) except MTurkRequestError as e: raise e def run(self): audio_file_dir = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/ind_trn" #audio_file_dir = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/dep_trn" prompt_file_uri = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/doc/al_sents.snr" base_clip_dir = "/home/taylor/data/corpora/LDC/LDC93S3A/rm_comp/rm1_audio1/rm1/clips" selection = 0 init_clip_count = 10000 while selection != "11": selection = raw_input("""Audio Source file to Audio Clip Approved Pipeline:\n 1: AudioSource-FileToClipped: Initialize Resource Management audio source files to %d queueable(Referenced) clips 2: AudioClip-ReferencedToHit: Queue all referenced audio clips and create a HIT if the queue is full. 3: AudioClip-HitToSubmitted: Check all submitted assignments for Transcriptions. 4: AudioClip-SubmittedToApproved: Check all submitted clips against their reference. 5: Review Current Hits 6: Worker liveness 7: Account balance 8: Worker stats 9: Recalculate worker WER 10: Assignment Stats 11: Exit """%init_clip_count) #selection = "5" if selection == "1": self._load_rm_audio_source_file_to_clipped(audio_file_dir, prompt_file_uri, base_clip_dir,init_clip_count=init_clip_count) elif selection == "2": self.audio_clip_referenced_to_hit() elif selection == "3": self.load_assignments_hit_to_submitted() elif selection == "4": self.assignment_submitted_approved() elif selection == "5": self.allhits_liveness() elif selection == "6": self.all_workers_liveness() elif selection == "7": print("Account balance: %s"%self.balance) elif selection == "8": self.stats() elif selection == "9": self.recalculate_worker_assignment_wer() elif selection == "10": self.get_assignment_stats() # def get_time_submitted_for_assignments(self): # assignments = self.mh.get_all_artifacts("assignments") # for assignment in assignments: # assignment_id = assignment["_id"] # a_assignment = self.conn.get_assignment(assignment_id)[0] # self.mh.update_artifact_by_id("assignments", assignment_id, "SubmitTime", a_assignment.SubmitTime) # def recalculate_worker_assignment_wer(self): # """For all submitted assignments, # if an answered question has a reference transcription, # check the WER. # If all the answered questions with reference transcriptions # have an acceptable WER, approve the assignment and update # the audio clips and transcriptions.""" # assignments = self.mh.get_artifacts("assignments",{"state":"Approved"}) # for assignment in assignments: # assignment_id = assignment["_id"] # denied = [] # #If no transcriptions have references then we automatically approve the HIT # approved = True # transcription_ids = assignment["transcriptions"] # transcriptions = self.mh.get_transcriptions("_id",transcription_ids) # worker_id = assignment["worker_id"] # worker_id = self.mh.create_worker_artifact(worker_id) # # max_rej_wer = (0.0,0.0) # total_wer = 0.0 # for transcription in transcriptions: # #Normalize the transcription # #self.mh.normalize_transcription # reference_id = self.mh.get_audio_clip_by_id(transcription["audio_clip_id"],"reference_transcription_id") # if reference_id: # reference_transcription = self.mh.get_reference_transcription({"_id": reference_id}, # "transcription") # new_transcription = transcription["transcription"].split(" ") # if reference_transcription: # transcription_wer = cer_wer(reference_transcription,new_transcription) # total_wer += transcription_wer # if transcription_wer < WER_THRESHOLD: # self.logger.info("WER for transcription(%s) %d"%(transcription["transcription"],transcription_wer)) # else: # max_rej_wer = (transcription_wer,WER_THRESHOLD) # denied.append((reference_transcription,new_transcription)) # approved = False # average_wer = total_wer/len(transcriptions) # #Update the worker # self.mh.add_assignment_to_worker(worker_id,(assignment_id,average_wer))
#HIT="2JNL8I9NZW6HG96GKYHWCT87ATCVL9" #HOST = 'mechanicalturk.amazonaws.com' # https://mechanicalturk.amazonaws.com/?Service=AWSMechanicalTurkRequester if SAND == 1: print "sandbox" HOST = 'mechanicalturk.sandbox.amazonaws.com' else: print "not Sandbox" HOST = 'mechanicalturk.amazonaws.com' conn = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) print conn.get_account_balance() def accept_and_pay(worker_id, assign_id, bonus_price, reason): """pays for assignment; returns False if something went wrong, else True""" try: result = conn.approve_assignment(assign_id) except MTurkRequestError: print "Couldn't pay" #TODO: make sure to avoid the possibility of paying the same bonus twice try: if bonus_price > 0: conn.grant_bonus(worker_id, assign_id, Price(amount=bonus_price), reason) except MTurkRequestError: #TODO: less embarrasing error handling
def Bank(): mtc = MTurkConnection(aws_access_key_id=ACCESS_ID, aws_secret_access_key=SECRET_KEY, host=HOST) print 'Funds: ' print mtc.get_account_balance()
parser = argparse.ArgumentParser( formatter_class=argparse.RawDescriptionHelpFormatter, description=""" Gets account balance at MTURK. The program reads the default config in the resources directory ('../resources/private/mturk.cfg') and any additional config files passed as an argument of a '-c'. The additional config file overwrites any default or previous values. Remember that the "private" directory is not distributed and it contains mainly information that you do not want to share. """) parser.add_argument('-c', "--configs", nargs='+', help='additional configuration files') args = parser.parse_args() mturk_cfg_fname = as_project_path('resources/private/mturk.cfg') cfg = Config.load_configs([mturk_cfg_fname] + args.configs, log=False) print "Gets MTURK account balance" print "-" * 120 print conn = MTurkConnection(aws_access_key_id = cfg['MTURK']['aws_access_key_id'], aws_secret_access_key = cfg['MTURK']['aws_secret_access_key'], host = cfg['MTURK']['host']) print "Account balance:", conn.get_account_balance()
Gets account balance at MTURK. The program reads the default config in the resources directory ('../resources/private/mturk.cfg') and any additional config files passed as an argument of a '-c'. The additional config file overwrites any default or previous values. Remember that the "private" directory is not distributed and it contains mainly information that you do not want to share. """) parser.add_argument('-c', "--configs", nargs='+', help='additional configuration files') args = parser.parse_args() mturk_cfg_fname = as_project_path('resources/private/mturk.cfg') cfg = Config.load_configs([mturk_cfg_fname] + args.configs, log=False) print "Gets MTURK account balance" print "-" * 120 print conn = MTurkConnection( aws_access_key_id=cfg['MTURK']['aws_access_key_id'], aws_secret_access_key=cfg['MTURK']['aws_secret_access_key'], host=cfg['MTURK']['host']) print "Account balance:", conn.get_account_balance()
class MTurkServices(object): ''' MTurk services ''' def __init__(self, aws_access_key_id, aws_secret_access_key, is_sandbox): self.update_credentials(aws_access_key_id, aws_secret_access_key) self.set_sandbox(is_sandbox) self.valid_login = self.verify_aws_login() if not self.valid_login: print 'WARNING *****************************' print 'Sorry, AWS Credentials invalid.\nYou will only be able to '\ 'test experiments locally until you enter\nvalid '\ 'credentials in the AWS Access section of ~/.psiturkconfig\n' def update_credentials(self, aws_access_key_id, aws_secret_access_key): ''' Update credentials ''' self.aws_access_key_id = aws_access_key_id self.aws_secret_access_key = aws_secret_access_key def set_sandbox(self, is_sandbox): ''' Set sandbox ''' self.is_sandbox = is_sandbox def get_reviewable_hits(self): ''' Get reviewable HITs ''' if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False reviewable_hits = [hit for hit in hits if hit.HITStatus == "Reviewable" \ or hit.HITStatus == "Reviewing"] hits_data = [MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration }) for hit in reviewable_hits] return hits_data def get_all_hits(self): ''' Get all HITs ''' if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False hits_data = [MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in hits] return hits_data def get_active_hits(self): ''' Get active HITs ''' if not self.connect_to_turk(): return False # hits = self.mtc.search_hits() try: hits = self.mtc.get_all_hits() except MTurkRequestError: return False active_hits = [hit for hit in hits if not hit.expired] hits_data = [MTurkHIT({ 'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsPending, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, }) for hit in active_hits] return hits_data def get_workers(self, assignment_status=None): ''' Get workers ''' if not self.connect_to_turk(): return False try: hits = self.mtc.get_all_hits() hit_ids = [hit.HITId for hit in hits] workers_nested = [] page_size=100 for hit_id in hit_ids: current_page_number=1 hit_assignments = self.mtc.get_assignments( hit_id, status=assignment_status, sort_by='SubmitTime', page_size=page_size, page_number=current_page_number ) totalNumResults = int(hit_assignments.TotalNumResults) total_pages = (totalNumResults // page_size) + (totalNumResults % page_size > 0) #do integer division then round up if necessary while current_page_number < total_pages: current_page_number += 1 hit_assignments += self.mtc.get_assignments( hit_id, status=assignment_status, sort_by='SubmitTime', page_size=page_size, page_number=current_page_number ) workers_nested.append(hit_assignments) workers = [val for subl in workers_nested for val in subl] # Flatten nested lists except MTurkRequestError: return False worker_data = [{ 'hitId': worker.HITId, 'assignmentId': worker.AssignmentId, 'workerId': worker.WorkerId, 'submit_time': worker.SubmitTime, 'accept_time': worker.AcceptTime, 'status': worker.AssignmentStatus } for worker in workers] return worker_data def bonus_worker(self, assignment_id, amount, reason=""): ''' Bonus worker ''' if not self.connect_to_turk(): return False try: bonus = MTurkConnection.get_price_as_price(amount) assignment = self.mtc.get_assignment(assignment_id)[0] worker_id = assignment.WorkerId self.mtc.grant_bonus(worker_id, assignment_id, bonus, reason) return True except MTurkRequestError as exception: print exception return False def approve_worker(self, assignment_id): ''' Approve worker ''' if not self.connect_to_turk(): return False try: self.mtc.approve_assignment(assignment_id, feedback=None) return True except MTurkRequestError: return False def reject_worker(self, assignment_id): ''' Reject worker ''' if not self.connect_to_turk(): return False try: self.mtc.reject_assignment(assignment_id, feedback=None) return True except MTurkRequestError: return False def unreject_worker(self, assignment_id): ''' Unreject worker ''' if not self.connect_to_turk(): return False try: self.mtc.approve_rejected_assignment(assignment_id) return True except MTurkRequestError: return False def verify_aws_login(self): ''' Verify AWS login ''' if ((self.aws_access_key_id == 'YourAccessKeyId') or (self.aws_secret_access_key == 'YourSecretAccessKey')): return False else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict( aws_access_key_id=self.aws_access_key_id, aws_secret_access_key=self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**mturkparams) try: self.mtc.get_account_balance() except MTurkRequestError as exception: print exception.error_message return False else: return True def connect_to_turk(self): ''' Connect to turk ''' if not self.valid_login: print 'Sorry, unable to connect to Amazon Mechanical Turk. AWS '\ 'credentials invalid.' return False if self.is_sandbox: host = 'mechanicalturk.sandbox.amazonaws.com' else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict( aws_access_key_id=self.aws_access_key_id, aws_secret_access_key=self.aws_secret_access_key, host=host) self.mtc = MTurkConnection(**mturkparams) return True def configure_hit(self, hit_config): ''' Configure HIT ''' # configure question_url based on the id experiment_portal_url = hit_config['ad_location'] frame_height = 600 mturk_question = ExternalQuestion(experiment_portal_url, frame_height) # Qualification: quals = Qualifications() approve_requirement = hit_config['approve_requirement'] quals.add( PercentAssignmentsApprovedRequirement("GreaterThanOrEqualTo", approve_requirement)) if hit_config['us_only']: quals.add(LocaleRequirement("EqualTo", "US")) # Create a HIT type for this HIT. hit_type = self.mtc.register_hit_type( hit_config['title'], hit_config['description'], hit_config['reward'], hit_config['duration'], keywords=hit_config['keywords'], approval_delay=None, qual_req=None)[0] # Check the config file to see if notifications are wanted. config = PsiturkConfig() config.load_config() try: url = config.get('Server Parameters', 'notification_url') all_event_types = [ "AssignmentAccepted", "AssignmentAbandoned", "AssignmentReturned", "AssignmentSubmitted", "HITReviewable", "HITExpired", ] self.mtc.set_rest_notification( hit_type.HITTypeId, url, event_types=all_event_types) except: pass # Specify all the HIT parameters self.param_dict = dict( hit_type=hit_type.HITTypeId, question=mturk_question, lifetime=hit_config['lifetime'], max_assignments=hit_config['max_assignments'], title=hit_config['title'], description=hit_config['description'], keywords=hit_config['keywords'], reward=hit_config['reward'], duration=hit_config['duration'], approval_delay=None, questions=None, qualifications=quals, response_groups=[ 'Minimal', 'HITDetail', 'HITQuestion', 'HITAssignmentSummary' ]) def check_balance(self): ''' Check balance ''' if not self.connect_to_turk(): return '-' return self.mtc.get_account_balance()[0] # TODO (if valid AWS credentials haven't been provided then # connect_to_turk() will fail, not error checking here and elsewhere) def create_hit(self, hit_config): ''' Create HIT ''' try: if not self.connect_to_turk(): return False self.configure_hit(hit_config) myhit = self.mtc.create_hit(**self.param_dict)[0] self.hitid = myhit.HITId except: return False else: return self.hitid # TODO(Jay): Have a wrapper around functions that serializes them. # Default output should not be serialized. def expire_hit(self, hitid): ''' Expire HIT ''' if not self.connect_to_turk(): return False try: self.mtc.expire_hit(hitid) return True except MTurkRequestError: print "Failed to expire HIT. Please check the ID and try again." return False def dispose_hit(self, hitid): ''' Dispose HIT ''' if not self.connect_to_turk(): return False try: self.mtc.dispose_hit(hitid) except Exception, e: print "Failed to dispose of HIT %s. Make sure there are no "\ "assignments remaining to be reviewed." % hitid
class MTurkProvider(object): description = 'This is a task authored by a requester on Daemo, a research crowdsourcing platform. ' \ 'Mechanical Turk workers are welcome to do it' keywords = ['daemo'] countries = ['US', 'CA'] min_hits = 1000 def __init__(self, host, aws_access_key_id, aws_secret_access_key): self.host = host self.connection = MTurkConnection( aws_access_key_id=aws_access_key_id, aws_secret_access_key=aws_secret_access_key, host=settings.MTURK_HOST ) self.connection.APIVersion = "2014-08-15" if not self.host: raise ValueError("Please provide a host url") def get_connection(self): return self.connection @staticmethod def _mturk_system_qualifications(qualification): requirements = [] for item in qualification.items.all(): if item.expression['attribute'] not in ['location', 'approval_rate', 'total_tasks']: continue requirement = None if item.expression['attribute'] == 'location': op = OP_IN if item.expression['operator'] == 'in' else OP_NOT_IN requirement = MultiLocaleRequirement(op, [val.strip() for val in item.expression['value'] if val is not None and val != '']) elif item.expression['attribute'] == 'approval_rate': op = OP_GT if item.expression['operator'] == 'gt' else OP_LT requirement = PercentAssignmentsApprovedRequirement(op, item.expression['value']) elif item.expression['attribute'] == 'total_tasks': op = OP_GT if item.expression['operator'] == 'gt' else OP_LT requirement = NumberHitsApprovedRequirement(op, item.expression['value']) requirements.append(requirement) return requirements def get_qualifications(self, project, boomerang_threshold, add_boomerang): requirements = [] if project.qualification is not None: requirements += self._mturk_system_qualifications(project.qualification) boomerang_qual, success = self.create_qualification_type(owner_id=project.owner_id, project_id=project.group_id, name='Boomerang Score #{}'.format(project.group_id), flag=FLAG_Q_BOOMERANG, description='No description available') boomerang = None if boomerang_threshold <= int(settings.BOOMERANG_MIDPOINT * 100): for i, bucket in enumerate(WAIT_LIST_BUCKETS): if int(bucket[1] * 100) <= boomerang_threshold: boomerang_blacklist, success = \ self.create_qualification_type(owner_id=project.owner_id, name='Boomerang Waitlist #{}-{}'.format(project.group_id, len( WAIT_LIST_BUCKETS) - i), flag=FLAG_Q_BOOMERANG, description='No description available', deny=True, project_id=project.group_id, bucket=bucket) if success and add_boomerang: boomerang = BoomerangRequirement(qualification_type_id=boomerang_blacklist.type_id, comparator=OP_DNE, integer_value=None) requirements.append(boomerang) else: boomerang = BoomerangRequirement(qualification_type_id=boomerang_qual.type_id, comparator=OP_GTEQ, integer_value=boomerang_threshold) if success and add_boomerang: requirements.append(boomerang) return Qualifications(requirements), boomerang_qual def create_hits(self, project, tasks=None, repetition=None): # if project.min_rating > 0: # return 'NOOP' if not tasks: cursor = connection.cursor() # noinspection SqlResolve query = ''' SELECT max(id) id, repetition, group_id, repetition - sum(existing_assignments) remaining_assignments, min_rating FROM ( SELECT t_rev.id, t.group_id, t.min_rating, p.repetition, CASE WHEN ma.id IS NULL OR ma.status IN (%(skipped)s, %(rejected)s, %(expired)s) THEN 0 ELSE 1 END existing_assignments FROM crowdsourcing_task t INNER JOIN crowdsourcing_project p ON t.project_id = p.id INNER JOIN crowdsourcing_task t_rev ON t_rev.group_id = t.group_id LEFT OUTER JOIN mturk_mturkhit mh ON mh.task_id = t_rev.id LEFT OUTER JOIN mturk_mturkassignment ma ON ma.hit_id = mh.id WHERE t.project_id = (%(project_id)s) AND t_rev.exclude_at IS NULL AND t_rev.deleted_at IS NULL ) t GROUP BY group_id, repetition, min_rating HAVING sum(existing_assignments) < repetition; ''' cursor.execute(query, {'skipped': TaskWorker.STATUS_SKIPPED, 'rejected': TaskWorker.STATUS_REJECTED, 'expired': TaskWorker.STATUS_EXPIRED, 'project_id': project.id}) tasks = cursor.fetchall() rated_workers = Rating.objects.filter(origin_type=Rating.RATING_REQUESTER).count() add_boomerang = rated_workers > 0 duration = project.timeout if project.timeout is not None else datetime.timedelta(hours=24) lifetime = project.deadline - timezone.now() if project.deadline is not None else datetime.timedelta( days=7) for task in tasks: question = self.create_external_question(task[0]) mturk_hit = MTurkHIT.objects.filter(task_id=task[0]).first() qualifications, boomerang_qual = self.get_qualifications(project=project, boomerang_threshold=int( round(task[4], 2) * 100), add_boomerang=add_boomerang) qualifications_mask = 0 if qualifications is not None: qualifications_mask = FLAG_Q_LOCALE + FLAG_Q_HITS + FLAG_Q_RATE + FLAG_Q_BOOMERANG hit_type, success = self.create_hit_type(title=project.name, description=self.description, price=project.price, duration=duration, keywords=self.keywords, approval_delay=datetime.timedelta(days=2), qual_req=qualifications, qualifications_mask=qualifications_mask, boomerang_threshold=int(round(task[4], 2) * 100), owner_id=project.owner_id, boomerang_qual=boomerang_qual) if not success: return 'FAILURE' if mturk_hit is None: try: hit = self.connection.create_hit(hit_type=hit_type.string_id, max_assignments=task[3], lifetime=lifetime, question=question)[0] self.set_notification(hit_type_id=hit.HITTypeId) mturk_hit = MTurkHIT(hit_id=hit.HITId, hit_type=hit_type, task_id=task[0]) except MTurkRequestError as e: error = e.errors[0][0] if error == 'AWS.MechanicalTurk.InsufficientFunds': message = { "type": "ERROR", "detail": "Insufficient funds on your Mechanical Turk account!", "code": error } redis_publisher = RedisPublisher(facility='bot', users=[project.owner]) message = RedisMessage(json.dumps(message)) redis_publisher.publish_message(message) return 'FAILED' else: if mturk_hit.hit_type_id != hit_type.id: result, success = self.change_hit_type_of_hit(hit_id=mturk_hit.hit_id, hit_type_id=hit_type.string_id) if success: mturk_hit.hit_type = hit_type mturk_hit.save() return 'SUCCESS' def create_hit_type(self, owner_id, title, description, price, duration, boomerang_threshold, keywords=None, approval_delay=None, qual_req=None, qualifications_mask=0, boomerang_qual=None): hit_type = MTurkHITType.objects.filter(owner_id=owner_id, name=title, description=description, price=Decimal(str(price)), duration=duration, qualifications_mask=qualifications_mask, boomerang_threshold=boomerang_threshold).first() if hit_type is not None: return hit_type, True reward = Price(price) try: mturk_ht = self.connection.register_hit_type(title=title, description=description, reward=reward, duration=duration, keywords=keywords, approval_delay=approval_delay, qual_req=qual_req)[0] hit_type = MTurkHITType(owner_id=owner_id, name=title, description=description, price=Decimal(str(price)), keywords=keywords, duration=duration, qualifications_mask=qualifications_mask, boomerang_qualification=boomerang_qual, boomerang_threshold=boomerang_threshold) hit_type.string_id = mturk_ht.HITTypeId hit_type.save() except MTurkRequestError: return None, False return hit_type, True def create_external_question(self, task, frame_height=800): task_hash = Hashids(salt=settings.SECRET_KEY, min_length=settings.ID_HASH_MIN_LENGTH) task_id = task_hash.encode(task) url = self.host + '/mturk/task/?taskId=' + task_id question = ExternalQuestion(external_url=url, frame_height=frame_height) return question def update_max_assignments(self, task): task = Task.objects.get(id=task['id']) mturk_hit = task.mturk_hit if not mturk_hit: raise MTurkHIT.DoesNotExist("This task is not associated to any mturk hit") assignments_completed = task.task_workers.filter(~Q(status__in=[TaskWorker.STATUS_REJECTED, TaskWorker.STATUS_SKIPPED, TaskWorker.STATUS_EXPIRED])).count() remaining_assignments = task.project.repetition - assignments_completed if remaining_assignments > 0 and mturk_hit.num_assignments == mturk_hit.mturk_assignments. \ filter(status=TaskWorker.STATUS_SUBMITTED).count() and \ mturk_hit.mturk_assignments.filter(status=TaskWorker.STATUS_IN_PROGRESS).count() == 0: self.add_assignments(hit_id=mturk_hit.hit_id, increment=1) self.extend_hit(hit_id=mturk_hit.hit_id) mturk_hit.status = MTurkHIT.STATUS_IN_PROGRESS mturk_hit.num_assignments += 1 mturk_hit.save() elif remaining_assignments == 0: self.expire_hit(hit_id=mturk_hit.hit_id) mturk_hit.status = MTurkHIT.STATUS_EXPIRED mturk_hit.save() elif remaining_assignments > 0 and \ mturk_hit.status == MTurkHIT.STATUS_EXPIRED: self.extend_hit(hit_id=mturk_hit.hit_id) mturk_hit.status = MTurkHIT.STATUS_IN_PROGRESS return 'SUCCESS' def get_assignment(self, assignment_id): try: return self.connection.get_assignment(assignment_id)[0], True except MTurkRequestError as e: error = e.errors[0][0] if error == 'AWS.MechanicalTurk.InvalidAssignmentState': return assignment_id, False return None, False def set_notification(self, hit_type_id): self.connection.set_rest_notification(hit_type=hit_type_id, url=self.host + '/api/mturk/notification', event_types=['AssignmentReturned', 'AssignmentAbandoned', 'AssignmentAccepted', 'AssignmentSubmitted']) def approve_assignment(self, task_worker): task_worker_obj = TaskWorker.objects.get(id=task_worker['id']) if hasattr(task_worker_obj, 'mturk_assignments') and task_worker_obj.mturk_assignments.first() is not None: try: self.connection.approve_assignment(task_worker_obj.mturk_assignments.first().assignment_id) except MTurkRequestError: return False return True def reject_assignment(self, task_worker): task_worker_obj = TaskWorker.objects.get(id=task_worker['id']) if hasattr(task_worker_obj, 'mturk_assignments') and task_worker_obj.mturk_assignments.first() is not None: try: self.connection.reject_assignment(task_worker_obj.mturk_assignments.first().assignment_id) except MTurkRequestError: return False return True def expire_hit(self, hit_id): try: self.connection.expire_hit(hit_id) except MTurkRequestError: return False return True def disable_hit(self, hit_id): try: self.connection.disable_hit(hit_id) except MTurkRequestError: return False return True def extend_hit(self, hit_id): try: self.connection.extend_hit(hit_id=hit_id, expiration_increment=604800) # 7 days except MTurkRequestError: return False return True def add_assignments(self, hit_id, increment=1): try: self.connection.extend_hit(hit_id=hit_id, assignments_increment=increment) except MTurkRequestError: return False return True def test_connection(self): try: return self.connection.get_account_balance()[0], True except MTurkRequestError as e: error = e.errors[0][0] if error == 'AWS.NotAuthorized': return None, False return None, False def get_account_balance(self): try: return self.connection.get_account_balance()[0] except MTurkRequestError: return None def create_qualification_type(self, owner_id, name, flag, description, project_id, auto_granted=False, auto_granted_value=None, deny=False, bucket=None): # noinspection SqlResolve query = ''' SELECT * FROM ( SELECT task.target_id, task.username, round(task.task_w_avg::NUMERIC, 2) rating --round(coalesce(task.task_w_avg, requester.requester_w_avg, -- platform.platform_w_avg)::NUMERIC, 2) rating FROM ( SELECT target_id, origin_id, project_id, username, sum(weight * power((%(BOOMERANG_TASK_ALPHA)s), t.row_number)) / sum(power((%(BOOMERANG_TASK_ALPHA)s), t.row_number)) task_w_avg FROM ( SELECT r.id, r.origin_id, p.group_id project_id, weight, r.target_id, -1 + row_number() OVER (PARTITION BY target_id ORDER BY tw.created_at DESC) AS row_number, u.username username FROM crowdsourcing_rating r INNER JOIN crowdsourcing_task t ON t.id = r.task_id INNER JOIN crowdsourcing_project p ON p.id = t.project_id INNER JOIN crowdsourcing_taskworker tw ON t.id = tw.task_id AND tw.worker_id=r.target_id INNER JOIN auth_user u ON u.id = r.target_id WHERE origin_id = (%(origin_id)s) AND origin_type = (%(origin_type)s)) t GROUP BY origin_id, target_id, project_id, username) task WHERE task.project_id = (%(project_id)s) ) r ''' extra_query = 'WHERE rating BETWEEN (%(lower_bound)s) AND (%(upper_bound)s);' params = { 'origin_type': Rating.RATING_REQUESTER, 'origin_id': owner_id, 'project_id': project_id, 'BOOMERANG_REQUESTER_ALPHA': settings.BOOMERANG_REQUESTER_ALPHA, 'BOOMERANG_PLATFORM_ALPHA': settings.BOOMERANG_PLATFORM_ALPHA, 'BOOMERANG_TASK_ALPHA': settings.BOOMERANG_TASK_ALPHA } obj_params = {'upper_bound': 300, 'lower_bound': 100} if deny and bucket is not None: query += extra_query params.update({'upper_bound': bucket[1], 'lower_bound': bucket[0]}) obj_params.update({'upper_bound': bucket[1] * 100, 'lower_bound': bucket[0] * 100, 'is_blacklist': True}) cursor = connection.cursor() cursor.execute(query, params=params) worker_ratings_raw = cursor.fetchall() worker_ratings = [{"worker_id": r[0], "worker_username": r[1], "rating": r[2]} for r in worker_ratings_raw] qualification = MTurkQualification.objects.filter(owner_id=owner_id, flag=flag, name=name).first() assigned_workers = [] if qualification is None: try: qualification_type = self.connection. \ create_qualification_type(name=name, description=description, status='Active', auto_granted=auto_granted, auto_granted_value=auto_granted_value)[0] qualification = MTurkQualification.objects.create(owner_id=owner_id, flag=flag, name=name, description=description, auto_granted=auto_granted, auto_granted_value=auto_granted_value, type_id=qualification_type.QualificationTypeId, **obj_params) except MTurkRequestError: return None, False else: assigned_workers = MTurkWorkerQualification.objects.values('worker').filter( qualification=qualification).values_list('worker', flat=True) for rating in worker_ratings: user_name = rating["worker_username"].split('.') if len(user_name) == 2 and user_name[0] == 'mturk': mturk_worker_id = user_name[1].upper() if mturk_worker_id not in assigned_workers: self.assign_qualification( qualification_type_id=qualification.type_id, worker_id=mturk_worker_id, value=int(rating['rating'] * 100)) defaults = { 'qualification': qualification, 'worker': mturk_worker_id, 'score': int(rating['rating'] * 100) } MTurkWorkerQualification.objects.update_or_create(qualification=qualification, worker=mturk_worker_id, defaults=defaults) return qualification, True def change_hit_type_of_hit(self, hit_id, hit_type_id): try: result = self.connection.change_hit_type_of_hit(hit_id=hit_id, hit_type=hit_type_id) except MTurkRequestError: return None, False return result, True def update_worker_boomerang(self, project_id, worker_id, task_avg, requester_avg): """ Update boomerang for project Args: project_id: worker_id: task_avg: requester_avg Returns: str """ hit = MTurkHIT.objects.select_related('hit_type__boomerang_qualification').filter( task__project__group_id=project_id).first() if hit is not None: qualification = hit.hit_type.boomerang_qualification worker_qual = MTurkWorkerQualification.objects.filter(qualification=qualification, worker=worker_id).first() if worker_qual is not None: self.update_score(worker_qual, score=int(task_avg * 100), override=True) else: MTurkWorkerQualification.objects.create(qualification=qualification, worker=worker_id, score=int(task_avg * 100), overwritten=True) self.assign_qualification(qualification_type_id=qualification.type_id, worker_id=worker_id, value=int(task_avg * 100)) # other_quals = MTurkWorkerQualification.objects.filter(~Q(qualification=qualification), # worker=worker_id, # overwritten=False) # for q in other_quals: # self.update_score(q, score=int(requester_avg * 100)) return 'SUCCESS' def update_score(self, worker_qual, score, override=False): if worker_qual is None: return False try: self.connection.update_qualification_score(worker_qual.qualification.type_id, worker_qual.worker, score) worker_qual.overwritten = override worker_qual.score = score worker_qual.save() except MTurkRequestError: return False return True def assign_qualification(self, qualification_type_id, worker_id, value=1): """ Revoke a qualification from a WorkerId Args: qualification_type_id: worker_id: value Returns: bool """ try: self.connection.assign_qualification(qualification_type_id, worker_id, value, send_notification=False) return True except MTurkRequestError: return False def revoke_qualification(self, qualification_type_id, worker_id): try: self.connection.revoke_qualification(qualification_type_id=qualification_type_id, subject_id=worker_id) return True except MTurkRequestError: return False def notify_workers(self, worker_ids, subject, message_text): try: self.connection.notify_workers(worker_ids, subject, message_text) return True except MTurkRequestError: return False
debug = True HOST = 'mechanicalturk.sandbox.amazonaws.com' MACYS_IMAGE_URL = 'http://i.imgur.com/1f7moUz.jpg' MACYS_IMAGE_TITLE = 'Material Girl Juniors Floral-Print Illusion Dress' PIN_IMAGE_URL = 'http://media-cache-ec0.pinimg.com/736x/de/21/31/de2131cb6d0fa52faaee161047194896.jpg' PIN_IMAGE_TITLE = 'Floral print maxi dress by Marc Jacobs.' mtc = MTurkConnection(aws_access_key_id=AWS_ACCESS_KEY_ID, aws_secret_access_key=AWS_SECRET_ACCESS_KEY, host=HOST) if debug: print mtc.get_account_balance() title = 'Match these Pictures to Macy\'s Products' description = 'Look at this photo and match it to Macy\'s products' keywords = 'clothing, rating, opinions, easy, quick, macys' ratings =[('Very Bad','1'), ('Bad','2'), ('OK','3'), ('Good','4'), ('Very Good','5')] #make overview overview = Overview() overview.append_field('Title', 'Rank how these two images match.')
class MTurkServices: def __init__(self, config): self.config = config def get_active_hits(self): self.connect_to_turk() # hits = self.mtc.search_hits() try: hits = self.mtc.get_all_hits() except MTurkRequestError: return(False) active_hits = [hit for hit in hits if not(hit.expired)] hits_data = [{'hitid': hit.HITId, 'title': hit.Title, 'status': hit.HITStatus, 'max_assignments': hit.MaxAssignments, 'number_assignments_completed': hit.NumberOfAssignmentsCompleted, 'number_assignments_pending': hit.NumberOfAssignmentsCompleted, 'number_assignments_available': hit.NumberOfAssignmentsAvailable, 'creation_time': hit.CreationTime, 'expiration': hit.Expiration, } for hit in active_hits] return(hits_data) def get_workers(self): self.connect_to_turk() try: hits = self.mtc.search_hits(sort_direction='Descending', page_size=20) hit_ids = [hit.HITId for hit in hits] workers_nested = [self.mtc.get_assignments( hit_id, status="Submitted", sort_by='SubmitTime', page_size=100 ) for hit_id in hit_ids] workers = [val for subl in workers_nested for val in subl] # Flatten nested lists except MTurkRequestError: return(False) completed_workers = [worker for worker in workers if worker.AssignmentStatus == "Submitted"] worker_data = [{'hitId': worker.HITId, 'assignmentId': worker.AssignmentId, 'workerId': worker.WorkerId, 'submit_time': worker.SubmitTime, 'accept_time': worker.AcceptTime } for worker in completed_workers] return(worker_data) def approve_worker(self, assignment_id): self.connect_to_turk() try: self.mtc.approve_assignment(assignment_id, feedback=None) except MTurkRequestError: return(False) def reject_worker(self, assignment_id): self.connect_to_turk() try: self.mtc.reject_assignment(assignment_id, feedback=None) except MTurkRequestError: return(False) def verify_aws_login(self, key_id, secret_key): is_sandbox = self.config.getboolean('HIT Configuration', 'using_sandbox') if is_sandbox: host = 'mechanicalturk.sandbox.amazonaws.com' else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict( aws_access_key_id=key_id, aws_secret_access_key=secret_key, host=host) self.mtc = MTurkConnection(**mturkparams) try: self.mtc.get_account_balance() except MTurkRequestError as e: print(e.error_message) print('AWS Credentials invalid') return 0 else: print('AWS Credentials valid') return 1 def connect_to_turk(self): is_sandbox = self.config.getboolean('HIT Configuration', 'using_sandbox') if is_sandbox: host = 'mechanicalturk.sandbox.amazonaws.com' else: host = 'mechanicalturk.amazonaws.com' mturkparams = dict( aws_access_key_id = self.config.get('AWS Access', 'aws_access_key_id'), aws_secret_access_key = self.config.get('AWS Access', 'aws_secret_access_key'), host=host) self.mtc = MTurkConnection(**mturkparams) def configure_hit(self): # Configure portal experimentPortalURL = self.config.get('HIT Configuration', 'question_url') frameheight = 600 mturkQuestion = ExternalQuestion(experimentPortalURL, frameheight) # Qualification: quals = Qualifications() approve_requirement = self.config.get('HIT Configuration', 'Approve_Requirement') quals.add( PercentAssignmentsApprovedRequirement("GreaterThanOrEqualTo", approve_requirement)) if self.config.getboolean('HIT Configuration', 'US_only'): quals.add(LocaleRequirement("EqualTo", "US")) # Specify all the HIT parameters self.paramdict = dict( hit_type = None, question = mturkQuestion, lifetime = datetime.timedelta(hours=self.config.getfloat('HIT Configuration', 'HIT_lifetime')), max_assignments = self.config.getint('HIT Configuration', 'max_assignments'), title = self.config.get('HIT Configuration', 'title'), description = self.config.get('HIT Configuration', 'description'), keywords = self.config.get('HIT Configuration', 'keywords'), reward = self.config.getfloat('HIT Configuration', 'reward'), duration = datetime.timedelta(hours=self.config.getfloat('HIT Configuration', 'duration')), approval_delay = None, questions = None, qualifications = quals ) def is_signed_up(self): access_key_id = self.config.get('AWS Access', 'aws_access_key_id') access_key = self.config.get('AWS Access', 'aws_secret_access_key') return (access_key_id != 'YourAccessKeyId') and \ (access_key != 'YourSecreteAccessKey') def check_balance(self): if self.is_signed_up(): self.connect_to_turk() return(self.mtc.get_account_balance()[0]) else: return('-') # TODO (if valid AWS credentials haven't been provided then connect_to_turk() will # fail, not error checking here and elsewhere) def create_hit(self): self.connect_to_turk() self.configure_hit() myhit = self.mtc.create_hit(**self.paramdict)[0] self.hitid = myhit.HITId # TODO(Jay): Have a wrapper around functions that serializes them. # Default output should not be serialized. def expire_hit(self, hitid): self.connect_to_turk() self.mtc.expire_hit(hitid) def extend_hit(self, hitid, assignments_increment=None, expiration_increment=None): self.connect_to_turk() self.mtc.extend_hit(hitid, assignments_increment=int(assignments_increment)) self.mtc.extend_hit(hitid, expiration_increment=int(expiration_increment)*60) def get_summary(self): try: balance = self.check_balance() summary = jsonify(balance=str(balance)) return(summary) except MTurkRequestError as e: print(e.error_message) return(False)