def _compute_user_stats(): """ Computes user statistics for the WMT13 evaluation campaign. """ user_stats = [] wmt13 = Group.objects.get(name='WMT13') users = wmt13.user_set.all() for user in users: _user_stats = HIT.compute_status_for_user(user) _name = user.username _avg_time = seconds_to_timedelta(_user_stats[1]) _total_time = seconds_to_timedelta(_user_stats[2]) _data = (_name, _user_stats[0], _avg_time, _total_time) if _data[0] > 0: user_stats.append(_data) # Sort by total number of completed HITs. user_stats.sort(key=lambda x: x[1]) user_stats.reverse() # Only show top 25 contributors. user_stats = user_stats[:25] return user_stats
def _compute_language_pair_stats(): """ Computes HIT statistics per language pair. """ language_pair_stats = [] for choice in LANGUAGE_PAIR_CHOICES: _code = choice[0] _name = choice[1] _remaining_hits = HIT.compute_remaining_hits(language_pair=_code) _total_hits = 0 _completed_hits = 0 for hit in HIT.objects.filter(active=True, mturk_only=False, language_pair=_code): _total_hits = _total_hits + 1 # Again: we now consider a HIT to be completed once it has been # annotated by one or more annotators. # # Before we required `hit.users.count() >= 3` for gr if hit.users.all().count() >= 1: _completed_hits = _completed_hits + 1 # _data = (_remaining_hits, _completed_hits, _total_hits) _data = ( _name, (_remaining_hits, 100 * _remaining_hits/float(_total_hits or 1)), (_completed_hits, 100 * _completed_hits/float(_total_hits or 1)) ) language_pair_stats.append(_data) return language_pair_stats
def status(request): """ Renders the status overview. """ LOGGER.info('Rendering WMT13 HIT status for user "{0}".'.format(request.user.username or "Anonymous")) # Compute some global stats. global_stats = [] wmt13 = Group.objects.get(name="WMT13") users = wmt13.user_set.all() # Check how many HITs have been completed. hits_completed = 0 for hit in HIT.objects.all(): if hit.users.count() >= 3: hits_completed = hits_completed + 1 # Compute remaining HITs for all language pairs. hits_remaining = HIT.compute_remaining_hits() # Compute number of results contributed so far. ranking_results = RankingResult.objects.all().count() # Aggregate information about participating groups. groups = set() for user in users: for group in user.groups.all(): if group.name == "WMT13" or group.name.startswith("eng2") or group.name.endswith("2eng"): continue groups.add(group) # Compute average/total duration over all results. durations = RankingResult.objects.all().values_list("duration", flat=True) total_time = sum([datetime_to_seconds(x) for x in durations]) avg_time = total_time / float(hits_completed or 1) global_stats.append(("Users", users.count())) global_stats.append(("Groups", len(groups))) global_stats.append(("HITs completed", hits_completed)) global_stats.append(("HITs remaining", hits_remaining)) global_stats.append(("Ranking results", ranking_results)) global_stats.append(("System comparisons", 10 * ranking_results)) global_stats.append(("Average duration", seconds_to_timedelta(avg_time))) global_stats.append(("Total duration", seconds_to_timedelta(total_time))) dictionary = { "active_page": "STATUS", "global_stats": global_stats, "commit_tag": COMMIT_TAG, "title": "WMT13 Status", } return render(request, "wmt13/status.html", dictionary)
def _compute_global_stats(): """ Computes some global statistics for the WMT13 evaluation campaign. """ global_stats = [] wmt13 = Group.objects.get(name='WMT13') users = wmt13.user_set.all() # Check how many HITs have been completed. We now consider a HIT to be # completed once it has been annotated by one or more annotators. # # Before we required `hit.users.count() >= 3` for greater overlap. hits_completed = 0 for hit in HIT.objects.filter(active=True, mturk_only=False): if hit.users.count() >= 1: hits_completed = hits_completed + 1 # Compute remaining HITs for all language pairs. hits_remaining = HIT.compute_remaining_hits() # Compute number of results contributed so far. ranking_results = RankingResult.objects.filter( item__hit__active=True, item__hit__mturk_only=False).count() # Aggregate information about participating groups. groups = set() for user in users: for group in user.groups.all(): if group.name == 'WMT13' or group.name.startswith('eng2') \ or group.name.endswith('2eng'): continue groups.add(group) # Compute average/total duration over all results. durations = RankingResult.objects.all().values_list('duration', flat=True) total_time = sum([datetime_to_seconds(x) for x in durations]) avg_time = total_time / float(hits_completed or 1) avg_user_time = total_time / float(3 * hits_completed or 1) global_stats.append(('Users', users.count())) global_stats.append(('Groups', len(groups))) global_stats.append(('HITs completed', hits_completed)) global_stats.append(('HITs remaining', hits_remaining)) global_stats.append(('Ranking results', ranking_results)) global_stats.append(('System comparisons', 10 * ranking_results)) global_stats.append(('Average duration', seconds_to_timedelta(avg_time))) global_stats.append(('Average duration (single user)', seconds_to_timedelta(avg_user_time))) global_stats.append(('Total duration', seconds_to_timedelta(total_time))) return global_stats
def _compute_group_stats(): """ Computes group statistics for the WMT13 evaluation campaign. """ group_stats = [] wmt13 = Group.objects.get(name='WMT13') users = wmt13.user_set.all() # Aggregate information about participating groups. groups = set() for user in users: for group in user.groups.all(): if group.name == 'WMT13' or group.name.startswith('eng2') \ or group.name.endswith('2eng'): continue groups.add(group) # The following dictionary defines the number of HITs each group should # have completed during the WMT13 evaluation campaign. group_hit_requirements = {'BALAGUR': 200, 'CMU': 300, 'CU': 1700, 'DCU': 300, 'DESRT': 100, 'FDA': 300, 'ITS-LATL': 200, 'JHU': 900, 'KIT': 400, 'LIA': 200, 'LIMSI': 600, 'MES': 1400, 'OMNIFLUENT': 400, 'Prompsit': 400, 'PROMT': 500, 'QUAERO': 100, 'RWTH': 400, 'SHEF': 700, 'STANFORD': 200, 'TALP': 100, 'TUBITAK': 200, 'UCAM': 100, 'UEDIN': 1700, 'UMD': 200, 'UU': 100, 'DFKI': 0, 'USAAR': 0} for group in groups: _name = group.name if not _name in group_hit_requirements.keys(): continue _group_stats = HIT.compute_status_for_group(group) _total = _group_stats[0] _required = group_hit_requirements[_name] _delta = _total - _required _data = (_total, _required, _delta) if _data[0] > 0: group_stats.append((_name, _data)) # Sort by number of remaining HITs. group_stats.sort(key=lambda x: x[1][2]) return group_stats
Author: Christian Federmann <*****@*****.**> usage: export_wmt13_status.py Exports HIT status for all language pairs. """ from datetime import datetime import os import sys if __name__ == "__main__": # Properly set DJANGO_SETTINGS_MODULE environment variable. os.environ['DJANGO_SETTINGS_MODULE'] = 'settings' PROJECT_HOME = os.path.normpath(os.getcwd() + "/..") sys.path.append(PROJECT_HOME) # We have just added appraise to the system path list, hence this works. from appraise.wmt13.models import HIT, LANGUAGE_PAIR_CHOICES remaining_hits = {} for language_pair in [x[0] for x in LANGUAGE_PAIR_CHOICES]: remaining_hits[language_pair] = HIT.compute_remaining_hits( language_pair=language_pair) print print '[{0}]'.format(datetime.now().strftime("%c")) for k, v in remaining_hits.items(): print '{0}: {1:03d}'.format(k, v) print
def overview(request): """ Renders the evaluation tasks overview. """ LOGGER.info('Rendering WMT13 HIT overview for user "{0}".'.format(request.user.username or "Anonymous")) # Re-initialise random number generator. seed(None) # Collect available language pairs for the current user. language_codes = set([x[0] for x in LANGUAGE_PAIR_CHOICES]) language_pairs = request.user.groups.filter(name__in=language_codes) hit_data = [] total = [0, 0, 0] for language_pair in language_pairs: hit = _compute_next_task_for_user(request.user, language_pair) status = HIT.compute_status_for_user(request.user, language_pair) for i in range(3): total[i] = total[i] + status[i] if hit: # Convert status seconds back into datetime.time instances. for i in range(2): status[i + 1] = seconds_to_timedelta(int(status[i + 1])) hit_data.append((hit.get_language_pair_display(), hit.get_absolute_url(), hit.block_id, status)) # Convert total seconds back into datetime.time instances. for i in range(2): total[i + 1] = seconds_to_timedelta(int(total[i + 1])) group = None for _group in request.user.groups.all(): if _group.name == "WMT13" or _group.name.startswith("eng2") or _group.name.endswith("2eng"): continue group = _group break if group is not None: group_name = group.name group_status = HIT.compute_status_for_group(group) for i in range(2): group_status[i + 1] = seconds_to_timedelta(int(group_status[i + 1])) else: group_status = None group_name = None LOGGER.debug( u'\n\nHIT data for user "{0}":\n\n{1}\n'.format( request.user.username or "Anonymous", u"\n".join([u"{0}\t{1}\t{2}\t{3}".format(*x) for x in hit_data]) ) ) dictionary = { "active_page": "OVERVIEW", "commit_tag": COMMIT_TAG, "hit_data": hit_data, "total": total, "group_name": group_name, "group_status": group_status, "title": "WMT13 Dashboard", } return render(request, "wmt13/overview.html", dictionary)
# Hotfix potentially wrong ISO codes; we are using ISO-639-3. iso_639_2_to_3_mapping = {'cze': 'ces', 'fre': 'fra', 'ger': 'deu'} for part2_code, part3_code in iso_639_2_to_3_mapping.items(): language_pair = language_pair.replace(part2_code, part3_code) try: _total = _total + 1 _hit_xml = tostring(_child, encoding="utf-8").decode('utf-8') if args.dry_run_enabled: _ = HIT(block_id=block_id, hit_xml=_hit_xml, language_pair=language_pair, mturk_only=args.mturk_only) else: # Use get_or_create() to avoid exact duplicates. We do allow # them for WMT13 to measure intra-annotator agreement... h = HIT(block_id=block_id, hit_xml=_hit_xml, language_pair=language_pair, mturk_only=args.mturk_only) h.save() # pylint: disable-msg=W0703 except Exception, msg: print msg _errors = _errors + 1 print print 'Successfully imported {0} HITs, encountered errors for ' \ '{1} HITs.'.format(_total, _errors) print
usage: export_wmt13_status.py Exports HIT status for all language pairs. """ from datetime import datetime import os import sys if __name__ == "__main__": # Properly set DJANGO_SETTINGS_MODULE environment variable. os.environ['DJANGO_SETTINGS_MODULE'] = 'settings' PROJECT_HOME = os.path.normpath(os.getcwd() + "/..") sys.path.append(PROJECT_HOME) # We have just added appraise to the system path list, hence this works. from appraise.wmt13.models import HIT, LANGUAGE_PAIR_CHOICES remaining_hits = {} for language_pair in [x[0] for x in LANGUAGE_PAIR_CHOICES]: remaining_hits[language_pair] = HIT.compute_remaining_hits( language_pair=language_pair) print print '[{0}]'.format(datetime.now().strftime("%c")) for k, v in remaining_hits.items(): print '{0}: {1:03d}'.format(k, v) print
def overview(request): """ Renders the evaluation tasks overview. """ LOGGER.info('Rendering WMT13 HIT overview for user "{0}".'.format( request.user.username or "Anonymous")) # Re-initialise random number generator. seed(None) # Collect available language pairs for the current user. language_codes = set([x[0] for x in LANGUAGE_PAIR_CHOICES]) language_pairs = request.user.groups.filter(name__in=language_codes) hit_data = [] total = [0, 0, 0] for language_pair in language_pairs: hit = _compute_next_task_for_user(request.user, language_pair) user_status = HIT.compute_status_for_user(request.user, language_pair) for i in range(3): total[i] = total[i] + user_status[i] if hit: # Convert status seconds back into datetime.time instances. for i in range(2): user_status[i+1] = seconds_to_timedelta(int(user_status[i+1])) hit_data.append( (hit.get_language_pair_display(), hit.get_absolute_url(), hit.block_id, user_status) ) # Convert total seconds back into datetime.timedelta instances. total[1] = seconds_to_timedelta(int(total[2]) / float(int(total[0]) or 1)) # Remove microseconds to get a nicer timedelta rendering in templates. total[1] = total[1] - timedelta(microseconds=total[1].microseconds) total[2] = seconds_to_timedelta(int(total[2])) group = None for _group in request.user.groups.all(): if _group.name == 'WMT13' \ or _group.name.startswith('eng2') \ or _group.name.endswith('2eng'): continue group = _group break if group is not None: group_name = group.name group_status = HIT.compute_status_for_group(group) for i in range(2): group_status[i+1] = seconds_to_timedelta(int(group_status[i+1])) else: group_status = None group_name = None LOGGER.debug(u'\n\nHIT data for user "{0}":\n\n{1}\n'.format( request.user.username or "Anonymous", u'\n'.join([u'{0}\t{1}\t{2}\t{3}'.format(*x) for x in hit_data]))) dictionary = { 'active_page': "OVERVIEW", 'commit_tag': COMMIT_TAG, 'hit_data': hit_data, 'total': total, 'group_name': group_name, 'group_status': group_status, 'title': 'WMT13 Dashboard', } return render(request, 'wmt13/overview.html', dictionary)
block_id = _child.attrib["block-id"] language_pair = '{0}2{1}'.format(_child.attrib["source-language"], _child.attrib["target-language"]) # Hotfix potentially wrong ISO codes; we are using ISO-639-3. iso_639_2_to_3_mapping = {'cze': 'ces', 'fre': 'fra', 'ger': 'deu'} for part2_code, part3_code in iso_639_2_to_3_mapping.items(): language_pair = language_pair.replace(part2_code, part3_code) try: _total = _total + 1 _hit_xml = tostring(_child, encoding="utf-8").decode('utf-8') if args.dry_run_enabled: _ = HIT(block_id=block_id, hit_xml=_hit_xml, language_pair=language_pair, mturk_only=args.mturk_only) else: # Use get_or_create() to avoid exact duplicates. We do allow # them for WMT13 to measure intra-annotator agreement... h = HIT(block_id=block_id, hit_xml=_hit_xml, language_pair=language_pair, mturk_only=args.mturk_only) h.save() # pylint: disable-msg=W0703 except Exception, msg: print msg _errors = _errors + 1