def random_forest_generator(data_filename, replacer,train_ratio = 0.8):
#initialize variables
data = pd.read_csv(data_filename)
data_labels = data['state'].tolist()
data_texts = data['content']
bow_transformer = CountVectorizer()
#vectorize train data
word_list = preprocess(data_texts, replacer)
vectorized_data = bow_transformer.fit_transform(word_list)
tfidf_transformer = TfidfTransformer().fit(vectorized_data)
worker = Worker(bow = bow_transformer,tfidf = tfidf_transformer)
vectorized_data_list = tfidf_transformer.transform(vectorized_data).toarray()
#print(vectorized_data_list)
all_data = worker.sliceData(vectorized_data_list,data_labels,train_ratio)
train_data = all_data['train_data']
train_labels = all_data['train_labels']
test_data = all_data['test_data']
test_labels = all_data['test_labels']
#build model
forest = RandomForestClassifier(n_estimators=30,n_jobs=-1)
forest = forest.fit(train_data,train_labels)
pred=forest.predict(test_data)
accuracy = worker.calculate_result(test_labels,pred)
#save data and model
joblib.dump(forest,'./var/model/forest_0',compress = 3)
joblib.dump(bow_transformer,'./var/model/bow_0',compress = 3)
joblib.dump(tfidf_transformer,'./var/model/tfidf_0', compress = 3)
return forest
def logistic_regression_generator(data_filename, replacer, train_ratio = 0.8):
data = pd.read_csv(data_filename)
data_labels = data['state'].tolist()
data_texts = data['content']
bow_transformer = CountVectorizer()
#vectorize train data
word_list = preprocess(data_texts, replacer)
vectorized_data = bow_transformer.fit_transform(word_list)
tfidf_transformer = TfidfTransformer().fit(vectorized_data)
worker = Worker(bow = bow_transformer,tfidf = tfidf_transformer)
vectorized_data_list = tfidf_transformer.transform(vectorized_data).toarray()
all_data = worker.sliceData(vectorized_data_list,data_labels,train_ratio)
train_data = all_data['train_data']
train_labels = all_data['train_labels']
test_data = all_data['test_data']
test_labels = all_data['test_labels']
#build model
lr = LogisticRegression(C=30,penalty='l2',random_state=0)
lr.fit(train_data,train_labels)
pred=lr.predict(test_data)
print("generator prediction:")
accuracy = worker.calculate_result(test_labels,pred)
#save data and model
joblib.dump(lr,'./var/model/lr_0',compress = 3)
joblib.dump(bow_transformer,'./var/model/bow_0',compress = 3)
joblib.dump(tfidf_transformer,'./var/model/tfidf_0', compress = 3)
return lr
def test_with_unlabeled_state(data_filename, replacer,classifier, bow, tfidf):
data = pd.read_csv(data_filename)
bow_transformer = joblib.load(bow)
tfidf_transformer = joblib.load(tfidf)
classifier = joblib.load(classifier)
data_texts = data['content']
worker = Worker(bow = bow_transformer,tfidf = tfidf_transformer)
word_list = preprocess(data_texts, replacer)
predict_data = worker.classifier_predict(word_list,classifier)
predict_data.to_csv('predict_data.csv')
def test_with_labeled_state(data_filename,replacer, classifier, bow,tfidf):
data = pd.read_csv(data_filename)
bow_transformer = joblib.load(bow)
tfidf_transformer = joblib.load(tfidf)
classifier = joblib.load(classifier)
data_labels = data['state'].tolist()
data_texts = data['content']
#preprocess data
worker = Worker(bow = bow_transformer,tfidf = tfidf_transformer)
word_list = preprocess(data_texts, replacer)
text_vectors_list = worker.content_to_vectors(word_list)
#test model
pred = classifier.predict(text_vectors_list)
worker.calculate_result(data_labels,pred)
def get_idle_worker():
doc = es.search_one_document(w_idx, es.missing_field_query("current_task_id"))
if doc:
worker = Worker.from_es_data(doc['_source'])
worker_id = doc['_id']
return worker, worker_id
return None, None
def main():
print "Content-type: text/html\n\n"
cgitb.enable()
# process input
form = cgi.FieldStorage()
xml_feedback = form.getfirst('xmlFeedback', 'None')
token_number = form.getfirst('token', 'None')
# open database
session = get_local_db_session()
try:
token = session.query(Token).filter_by(number=token_number).one()
except NoResultFound:
print "KO"
return
# insert the retrived dialogueID into the xmlFeedback
xml_feedback = xml_feedback.replace(
"<dialogueId></dialogueId>",
"<dialogueId>" + token.submission.dialogue_id + "</dialogueId>")
worker_id = get_worker_id(xml_feedback)
phone = token.submission.dialogue_id.rsplit("-", 1)[1]
try:
curr_worker = session.query(Worker).filter_by(
worker_id=worker_id).one()
except NoResultFound:
curr_worker = Worker(worker_id=worker_id, phone_number=phone)
session.add(curr_worker)
# save submitted data
submission = token.submission
submission.data = xml_feedback
submission.worker = curr_worker
submission.timestamp = datetime.datetime.now()
# free token
token.submission = None
session.commit()
def job(csv_filename=csv_filename, bow_transformer=bow_transformer, classifier=classifier, tfidf_transformer=tfidf_transformer, host=host, passcode=passcode, charset=charset, usr=usr, db=db,time_storage=time_storage):
#Start to load synonym pairs to SynonymReplacer
logger.info("Start to load synonym pairs to SynonymReplacer:")
try:
replacer = SynonymReplacer()
replacer.addSynDict_from_csv(csv_filename)
logger.info("Successfully load synonym pairs to SynonymReplacer.")
except Exception as err:
logger.error("Fail to load synonym pairs to SynonymReplacer. {}".format(err))
#replacer.show_syndict()
#forest = random_forest_generator('./var/model/data.csv',replacer)
#test_with_labeled_state('./var/model/model_data.csv',replacer,'./var/model/forest_0','./var/model/bow_0','./var/model/tfidf_0')
error_flag = False
logger.info("Start to load lasttime from time_storage.txt:")
try:
with open(time_storage,'r') as file:
lasttime = file.readline()
logger.info("Successfully load lasttime from time_storage.txt.")
except Exception as err:
lasttime = time.strftime( '%Y-%m-%d %X', time.localtime())
logger.error("Fail to load lasttime from time_storage.txt. Instead, set current time to be lasttime. {}".format(err))
#Connect to the database
logger.info("Start to connect to the database:")
try:
connection = pymysql.connect(host=host,
user= usr,
password= passcode,
db= db,
charset=charset,
cursorclass=pymysql.cursors.DictCursor)
logger.info("Successfully connect to the database.")
with connection.cursor() as cursor:
#retrieve data from the database
newtime = str(time.strftime( '%Y-%m-%d %X', time.localtime()))
flag = cursor.execute("SELECT * FROM GC_User_Comment WHERE CreatedTime >= %s and CreatedTime < %s",(lasttime,newtime))
if flag:
Recentdata = pd.DataFrame(cursor.fetchall())
Recentdata = Recentdata[['CMTID','Content','State','AuditState']]
logger.info("Successfully retrieve {} comments from database.".format(len(Recentdata)))
logger.info("Start to load models:")
try:
classifier = joblib.load(classifier)
bow_transformer = joblib.load(bow_transformer)
tfidf_transformer = joblib.load(tfidf_transformer)
logger.info("Successfully load models.")
except Exception as err:
logger.error("Fail to load models. {}".format(err))
error_flag = True
logger.info("Start to preprocecss Recentdata:")
try:
data_texts = Recentdata['Content']
word_list = preprocess(data_texts, replacer)
logger.info("Successfully preprocess Recentdata.")
except Exception as err:
logger.error("Fail to preprocess Recentdata. {}".format(err))
error_flag = True
logger.info("Start to predict Recentdata:")
try:
worker = Worker(bow = bow_transformer,tfidf = tfidf_transformer)
predict_data = worker.classifier_predict(word_list,classifier)
predict_data.to_csv('predict_data.csv')
logger.info("Successfully predict Recentdata")
Recentdata['State'] = [state_classify(prob) for prob in predict_data['trash']]
Recentdata['AuditState'] = [auditstate_classify(prob) for prob in predict_data['trash']]
except Exception as err:
logger.error("Fail to predict Recentdata. {}".format(err))
error_flag = True
logger.info("Start to update State and AuditState in the database:")
try:
count = 0
for CMTID, Content, State, AuditState in Recentdata.values:
cursor.execute("UPDATE GC_User_Comment SET State = %s, AuditState = %s WHERE CMTID = %s",(int(State), int(AuditState), int(CMTID)))
count = count + 1
connection.commit()
except Exception as err:
logger.error("Fail to update GC_User_Comment State. {0}".format(err))
error_flag = True
if(not error_flag):
#update lasttime into txt file
logger.info("Start to update time in time_storage.txt:")
try:
with open(time_storage,'w') as file:
file.writelines(newtime)
logger.info("Successfully updated time in time_storage.txt")
logger.info("Successfully updated State and AuditState for {} comment".format(count))
except Exception as err:
logger.error("Fail to update time in time_storage.txt. {}".format(err))
else:
logger.info("No new comment was created from the last recorded CreatedTime")
connection.close()
except Exception as err:
logger.error("Fail to connect to the database. {}".format(err))
def register_worker(platform_info, environment):
return db.register_worker(Worker(platform_info, environment))
def get_worker_by_id(worker_id):
doc = es.get_document_by_id(w_idx, worker_id)
if doc:
return Worker.from_es_data(doc['_source'])
return None
import torch as th
from utils import SharedAdam
from model import Worker, Net
import gym
import torch.multiprocessing as mp
env = gym.make('CartPole-v0')
n_action = env.action_space.n
n_state = env.observation_space.shape[0]
global_net = Net(n_state, n_action)
global_net.share_memory()
optA = SharedAdam(global_net.policy.parameters(), lr=1e-4, betas=(0.92, 0.999))
optC = SharedAdam(global_net.v.parameters(), lr=1e-4, betas=(0.92, 0.999))
workers = [Worker(global_net, optA, optC, str(i)) for i in range(8)]
[w.start() for w in workers]
[w.join() for w in workers]
