def DoTest(classA,
classB,
unsupervised=False,
balance=False,
use_baseline_segmenter=False):
classA = list(classA)
classB = list(classB)
if balance:
max_len = min(len(classA), len(classB))
classA = classA[:max_len]
classB = classB[:max_len]
random.shuffle(classA)
random.shuffle(classB)
percents = (0.2, 0.1, 0.7)
classA_test, classA_validation, classA_train = Partition(classA, percents)
classB_test, classB_validation, classB_train = Partition(classB, percents)
if use_baseline_segmenter:
seg_func = segmenter.baseline_segmenter
else:
seg_func = segmenter.morph_segmenter(model)
classifier = BinaryClassifier.Train(seg_func, classA_train, classB_train)
classifier.GetTopRatios()
thresh = GetOptimalThreshold(classifier, classA_validation,
classB_validation)
acc = TestAccuracy(classifier, classA_test, classB_test, thresh)
print 'test accuracy {0}'.format(acc)
if unsupervised:
semisup_classifier = classifier
for iter_num in range(3):
print 'semi-sup iter {0}'.format(iter_num)
semisup_classifier.TrainConfidenceEstimator(
classA_validation, classB_validation)
semisup_classifier = BinaryClassifier.TrainSemiSupervised(
snapchat_names, semisup_classifier)
thresh = GetOptimalThreshold(semisup_classifier, classA_validation,
classB_validation)
acc = TestAccuracy(semisup_classifier, classA_test, classB_test,
thresh)
print 'accuracy {0}'.format(acc)
def DoTest(classA, classB, unsupervised=False, balance=False,
use_baseline_segmenter=False):
classA = list(classA)
classB = list(classB)
if balance:
max_len = min(len(classA), len(classB))
classA = classA[:max_len]
classB = classB[:max_len]
random.shuffle(classA)
random.shuffle(classB)
percents = (0.2, 0.1, 0.7)
classA_test, classA_validation, classA_train = Partition(classA, percents)
classB_test, classB_validation, classB_train = Partition(classB, percents)
if use_baseline_segmenter:
seg_func = segmenter.baseline_segmenter
else:
seg_func = segmenter.morph_segmenter(model)
classifier = BinaryClassifier.Train(seg_func, classA_train, classB_train)
classifier.GetTopRatios()
thresh = GetOptimalThreshold(classifier,
classA_validation, classB_validation)
acc = TestAccuracy(classifier, classA_test, classB_test, thresh)
print 'test accuracy {0}'.format(acc)
if unsupervised:
semisup_classifier = classifier
for iter_num in range(3):
print 'semi-sup iter {0}'.format(iter_num)
semisup_classifier.TrainConfidenceEstimator(classA_validation, classB_validation)
semisup_classifier = BinaryClassifier.TrainSemiSupervised(snapchat_names,
semisup_classifier)
thresh = GetOptimalThreshold(semisup_classifier, classA_validation, classB_validation)
acc = TestAccuracy(semisup_classifier, classA_test, classB_test, thresh)
print 'accuracy {0}'.format(acc)
acc = TestAccuracy(semisup_classifier, classA_test, classB_test, thresh)
print 'accuracy {0}'.format(acc)
segfun = semisup_classifier.segfun
with open("../models/init_nonum_semi%i.pkl" % (iter_num + 1), 'w') as f:
semisup_classifier.segfun = None
cPickle.dump(semisup_classifier, f)
semisup_classifier.segfun = segfun
return semisup_classifier
if __name__ == '__main__':
target = 'gender'
whereclause = "where gender is not ''" if target == 'gender' else ''
model = segmenter.load_model('../models/idmorphs_naworl.model')
segfun = segmenter.morph_segmenter(model, match='[a-z]+')
# model_semi = segmenter.load_model('../models/idmorphs.model')
# segfun_semi = segmenter.morph_segmenter(model_semi, match='[a-z]+')
users = segmenter.get_users_from_db(whereclause=whereclause)
male_ids = [user.id for user in users if user.gender == 'M']
female_ids = [user.id for user in users if user.gender == 'F']
# unlabeled_users = segmenter.get_users_from_db(tablename='naver')
# unknown_ids = [user.id for user in unlabeled_users]
unknown_ids = None
cls = DoTest(male_ids, female_ids, segfun, unknown_ids, balance=False)
# cls = cPickle.load(open("../models/init_nonum_semi3.pkl"))
# cls.segfun = segfun_semi
result = classifier.Classify(name)
result['lang'] = lang
result['name'] = name
results.append(result)
return pandas.DataFrame(results)
def get_preds(baseline, morph, weight):
columns = numpy.array(['True', 'False'])
z = weight * baseline[columns] + (1.0 - weight) * morph[columns]
idx = z.values.argmax(axis=1)
return columns[idx]
base_segmenter = segmenter.baseline_segmenter
morph_segmenter = segmenter.morph_segmenter(Classifier.model)
def getMetrics(truelabels, predlabels):
prec = metrics.precision_score(truelabels, predlabels, pos_label='True')
recall = metrics.recall_score(truelabels, predlabels, pos_label='True')
return prec, recall
all_langs = train.lang.unique()
for lang in all_langs:
labels = [str(x) for x in train.lang == lang]
testlabels = [str(x) for x in test.lang == lang]
baseline_classifier = BayesClassifier.Train(base_segmenter,
train.name_lower,
labels)
morph_classifier = BayesClassifier.Train(morph_segmenter,
result = classifier.Classify(name)
result['lang'] = lang
result['name'] = name
results.append(result)
return pandas.DataFrame(results)
def get_preds(baseline, morph, weight):
columns = numpy.array(['True', 'False'])
z = weight * baseline[columns] + (1.0 - weight) * morph[columns]
idx = z.values.argmax(axis=1)
return columns[idx]
base_segmenter = segmenter.baseline_segmenter
morph_segmenter = segmenter.morph_segmenter(Classifier.model)
def getMetrics(truelabels, predlabels):
prec = metrics.precision_score(truelabels, predlabels, pos_label='True')
recall = metrics.recall_score(truelabels, predlabels, pos_label='True')
return prec, recall
all_langs = train.lang.unique()
for lang in all_langs:
labels = [str(x) for x in train.lang == lang]
testlabels = [str(x) for x in test.lang == lang]
baseline_classifier = BayesClassifier.Train(base_segmenter,
train.name_lower, labels)
morph_classifier = BayesClassifier.Train(morph_segmenter, train.name_lower,