def connect(self):
if self._connected:
return
self._db = create_engine(self._connection_url)
with timeit('connect db %s' % self._debug_connection_url, 1000):
self._db.connect()
self._connected = True
type=int,
help='Test the tagger with TEST known examples (default 500)')
parser.add_argument('--test-mediahaven',
action='store_true',
help='Tag MediaHaven newspaper OCR-text')
parser.add_argument('--profile',
action='store_true',
help='Output run times of some key operations')
parser.add_argument(dest='pickle', help='Filename of pickle file')
args = parser.parse_args()
if not args.profile:
logging.getLogger('pythonmodules.profiling').setLevel(logging.ERROR)
samples = Samples(GMB())
if args.train:
with timeit('Creating NamedEntityChunker'):
chunker = NamedEntityChunker(samples.training())
pickle.dump(chunker, open(args.pickle, 'wb'))
else:
with timeit('Pickle load'):
chunker = pickle.load(open(args.pickle, 'rb'))
if args.test_mediahaven:
with timeit('NER Tagging'):
from pythonmodules.mediahaven import MediaHaven
# from pythonmodules.config import Config
mh = MediaHaven()
item = mh.one('+(workflow:GMS) +(archiveStatus:on_tape)')
print(chunker.parse(pos_tag(word_tokenize(item['description']))))
help='Extra where clause to pass to the select query (eg. "status=1")')
args = parser.parse_args()
logging.basicConfig()
logger = logging.getLogger()
fh = logging.FileHandler(args.log_file)
fh.setLevel(logging.INFO)
logger.addHandler(fh)
_solr = Solr(Config(section='solr')['url'])
config = Config(section='db')
conn = psycopg2.connect(config['connection_url'])
conn2 = psycopg2.connect(config['connection_url'])
cur = conn.cursor()
with timeit('SELECT', 5000):
where_clause = ('WHERE %s' % args.where) if args.where else ''
q = 'SELECT pid, nmlid, entity, id, score FROM %s %s ORDER BY pid ASC' % (
args.table, where_clause)
if args.limit:
q += ' LIMIT %d' % int(args.limit)
if args.start:
q += ' OFFSET %d' % int(args.start)
cur.execute(q)
@multithreaded(10, pre_start=True, pass_thread_id=False)
def process(row):
try:
res = _solr.search('id:%s' % row[0], rows=1, fl=['text', 'language'])
def process_pid(row):
try:
id_, full_pid, external_id, entity, score, meta = row
entity = remove_double_spaces.sub(' ', entity)
with timeit('Rater init', 1e3):
rater = Rater(full_pid, external_id, entity)
cur_rating = score
meta_old = None
if meta:
meta_old = meta
try:
meta = get_meta(full_pid, external_id, entity, score, meta)
except KeyError as e:
logger.warning(e)
with conn.cursor() as cur2:
cur2.execute(
'UPDATE ' + args.table + ' SET status=4 WHERE id=%s',
[id_])
# conn.commit()
new_score = 0
try:
rating = rater.ratings()
meta['rating_breakdown'] = {
k: rating.scores[k].rating
for k in rating.scores
}
meta['rating_multiplier'] = rating.total_multiplier
new_score = rating.total
except KeyError as e:
logger.warning(e)
toset = dict()
if not isclose(cur_rating, new_score, abs_tol=.0001):
toset['score'] = new_score
if 'quality' in meta:
del meta['quality']
meta = json.dumps(meta)
if meta_to_comp_meta(meta) != meta_to_comp_meta(meta_old):
toset['meta'] = meta
if not len(toset):
return
with timeit('SLOW UPDATE %s' % id_, 1e3), conn.cursor() as cur2:
keys = ','.join([k + ' = %s' for k in toset.keys()])
values = [toset[k] for k in toset.keys()]
values.append(id_)
cur2.execute(
'UPDATE ' + args.table + ' SET ' + keys + ' WHERE id=%s',
values)
# conn2.commit()
except Exception as e:
try:
url = 'http://do-tst-mke-01.do.viaa.be/attestation/info/model-%s/%s/%s/%s' % \
(model_name, full_pid, external_id, entity.replace(' ', '/'))
except Exception as e2:
url = str(e2)
logger.warning('exception for %s', url)
logger.exception(e)
def connect(self):
super().connect()
with timeit('reflect db %s' % self._debug_connection_url, 1000):
self._meta = MetaData(bind=self._db)
self._meta.reflect()
def test(self, amount=500):
# known_tags = NER.allowed_tags
samples = self.corpus.read_entities()
if amount == 0:
logger.info("Loading the entire corpus in memory, this may take a while...")
# preload all samples to have an amount available for progress indicator
samples = list(samples)
amount = len(samples)
totals = [defaultdict(lambda: 0) for i in range(len(self.taggers))]
# stats = [defaultdict(lambda: defaultdict(lambda: 0)) for i in range(len(self.taggers))]
total_tags = 0
progress = tqdm(total=amount*len(self.taggers))
timer = timeit()
full_orig_tags = []
full_predict_tags = [[] for i in range(len(self.taggers))]
for sample_index, sample in enumerate(samples):
if amount and sample_index >= amount:
break
sample_tags = Tester.filter_tags(sample.entities)
orig_tags = [tag[1] for tag in sample_tags]
ntags = len(orig_tags)
total_tags += ntags
full_orig_tags.extend(orig_tags)
for n, tagger in enumerate(self.taggers):
progress.update()
cls = type(tagger).__name__
timer.restart()
tags = Tester.filter_tags(tagger.tag(sample.phrase))
elapsed = timer.elapsed()
# check and do a naive attempt to fix different tag lengths
if len(tags) != ntags:
tags = Tester.fix_tags_counts(tags, sample_tags)
if len(tags) != ntags:
logger.error("Samples are of different size for %s (%d vs %d): \nTAGGER: %s\nORIGINAL: %s",
cls, len(tags), ntags, tags, sample_tags)
continue
tag_types = [tag[1] for tag in tags]
zipped = list(zip(tag_types, orig_tags))
sames = sum([tag[0] == tag[1] for tag in zipped])
# for tags in zipped:
# if tags[0] in known_tags:
# if tags[0] == tags[1]:
# stats[n][tags[0]]['tp'] += 1
# else:
# stats[n][tags[0]]['fp'] += 1
# elif tags[1] in known_tags:
# stats[n][tags[1]]['fn'] += 1
#
# for tag in known_tags:
# if tag != tags[1]:
# stats[n][tag]['tn'] += 1
totals[n]['same'] += sames
totals[n]['time'] += elapsed
full_predict_tags[n].extend(tag_types)
Stats = namedtuple('Stats', ['accuracy', 'time', 'total_checked', 'confusion_matrix'])
return [Stats(
t['same'] / total_tags * 100,
t['time'],
# stats[i],
total_tags,
ConfusionMatrix(actual_vector=full_orig_tags, predict_vector=full_predict_tags[i])
) for i, t in enumerate(totals)]
def clear(self):
with timeit('Clearing model %s' % self.link._meta.db_table):
# this doesnt reset identity and is slower...
# self.link.objects.all().delete()
self.db.execute('TRUNCATE TABLE %s RESTART IDENTITY' %
self.link._meta.db_table)
'Migrate statuses between attestation_link tables (doesn\'t overwrite if a status already set)'
)
parser.add_argument('from_table', help='Origin table')
parser.add_argument('to_table', help='Target table')
parser.add_argument('--debug',
action='store_true',
default=False,
help='Show debug log messages')
args = parser.parse_args()
logging.basicConfig(level=logging.DEBUG if args.debug else logging.INFO)
config = Config(section='db')
conn = psycopg2.connect(config['connection_url'])
cur = conn.cursor()
cur2 = conn.cursor()
with timeit('SELECT', 5000):
cur.execute(
'select status, kind, extras, pid, nmlid from %s where status != 0 AND status != 4'
% args.from_table)
processed = 0
total = cur.rowcount
q = 'UPDATE ' + args.to_table + ' SET status=%s, kind=%s, extras=%s WHERE pid=%s AND nmlid=%s AND status=0'
for idx, row in enumerate(tqdm(cur, total=total)):
res = cur2.execute(q, row)
processed += cur2.rowcount
logger.debug('%d changed for: %s', cur2.rowcount, row)
if idx % 10 == 0:
with timeit('commit', 250):
conn.commit()
