def test_get_one_metric(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'MultinomialNB'
metric = 'f1'
cf_cv.train_save_metrics([
('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB(alpha=.05)),
], metric, name, self.test_dir, self.test_dir)
f1 = cf_cv.get_one_metric_cv('f1')
precision = cf_cv.get_one_metric_cv('precision')
recall = cf_cv.get_one_metric_cv('recall')
support = cf_cv.get_one_metric_cv('support')
f1_average = cf_cv.get_one_metric_cv('f1', average=True)
precision_average = cf_cv.get_one_metric_cv('precision', average=True)
recall_average = cf_cv.get_one_metric_cv('recall', average=True)
support_average = cf_cv.get_one_metric_cv('support', average=True)
self.assertEqual(type(f1), pd.DataFrame)
self.assertEqual(type(precision), pd.DataFrame)
self.assertEqual(type(recall), pd.DataFrame)
self.assertEqual(type(support), pd.DataFrame)
self.assertEqual(type(support_average), pd.DataFrame)
self.assertEqual(type(recall_average), pd.DataFrame)
self.assertEqual(type(precision_average), pd.DataFrame)
self.assertEqual(type(f1_average), pd.DataFrame)
def test_ft_weights(self):
clf1 = Pipeline([('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', LogisticRegression())])
clf2 = Pipeline([('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB())])
clf3 = FasttextClassifier(epoch=2, output=self.ft_output)
eclf = EnsembleClassifier(clfs=[clf1, clf2, clf3], weights=[1,1,0.5])
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'ensemble_clf'
metric = 'f1'
cf_cv.train_save_metrics([('clf', eclf)],
metric, name,
self.test_dir,
self.test_dir)
self.assertTrue(os.path.isfile(os.path.join(self.test_dir, name + '_' + metric + '.png')))
self.assertTrue(os.path.isfile(os.path.join(self.test_dir, name + 'ROC_AUC.png')))
self.assertTrue(os.path.isfile(os.path.join(self.test_dir, name + 'prec_recall.png')))
self.assertTrue(os.path.isfile(os.path.join(self.test_dir, name + '.xlsx')))
self.assertTrue(os.path.isfile(os.path.join(self.test_dir, name + '_average.xlsx')))
self.assertEqual(type(cf_cv.roc_auc), dict)
self.assertEqual(type(cf_cv.tpr), dict)
self.assertEqual(type(cf_cv.fpr), dict)
self.assertEqual(type(cf_cv.metrics_average_df), pd.DataFrame)
self.assertEqual(type(cf_cv.metrics_df), pd.DataFrame)
self.assertEqual(type(cf_cv.metrics_per_class), list)
self.assertEqual(type(cf_cv.metrics_average), list)
def test_train_save_metrics(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'MultinomialNB'
metric = 'f1'
cf_cv.train_save_metrics([
('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB(alpha=.05)),
], metric, name, self.test_dir, self.test_dir)
self.assertTrue(
os.path.isfile(
os.path.join(self.test_dir, name + '_' + metric + '.png')))
self.assertTrue(
os.path.isfile(os.path.join(self.test_dir, name + 'ROC_AUC.png')))
self.assertTrue(
os.path.isfile(
os.path.join(self.test_dir, name + 'prec_recall.png')))
self.assertTrue(
os.path.isfile(os.path.join(self.test_dir, name + '.xlsx')))
self.assertTrue(
os.path.isfile(os.path.join(self.test_dir,
name + '_average.xlsx')))
self.assertEqual(type(cf_cv.roc_auc), dict)
self.assertEqual(type(cf_cv.tpr), dict)
self.assertEqual(type(cf_cv.fpr), dict)
self.assertEqual(type(cf_cv.metrics_average_df), pd.DataFrame)
self.assertEqual(type(cf_cv.metrics_df), pd.DataFrame)
self.assertEqual(type(cf_cv.metrics_per_class), list)
self.assertEqual(type(cf_cv.metrics_average), list)
def test_pickle(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'MultinomialNB'
metric = 'f1'
cf_cv.train_save_metrics([('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB(alpha=.05))], metric,
name, self.test_dir, self.test_dir)
savefile = os.path.join(self.test_dir, 'clf_cv.cv')
cf_cv.pickle(savefile)
self.assertTrue(os.path.isfile(savefile))
def test_predict_labels(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'MultinomialNB'
metric = 'f1'
cf_cv.train_save_metrics([
('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB(alpha=.05)),
], metric, name, self.test_dir, self.test_dir)
labels = cf_cv.predict(['bad', 'good'])
self.assertTrue(all(labels == ['neg', 'pos']))
def test_make_roc_auc_plot(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'MultinomialNB'
metric = 'f1'
cf_cv.train_save_metrics([('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB(alpha=.05))], metric,
name, self.test_dir, self.test_dir)
filename = os.path.join(self.test_dir, name + '_roc_auc_plot.png')
cf_cv.make_roc_auc_plot(savefile=filename)
self.assertTrue(os.path.isfile(filename))
def test_xft_classifiercv(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'ft'
metric = 'f1'
cf_cv.train_save_metrics(
[('clf', FasttextClassifier(output=self.output, epoch=1))], metric,
name, self.test_dir, self.test_dir)
filename = os.path.join(self.test_dir, '_eval_report')
cf_cv.calc_evaluation_report(self.df_test['text'],
self.df_test['class'],
savefile=filename)
self.assertTrue(os.path.isfile(filename + "_" + name + ".csv"))
self.assertTrue(os.path.isfile(filename + "_" + name + "_average.csv"))
def test_predict_labels_probas(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'MultinomialNB'
metric = 'f1'
cf_cv.train_save_metrics([
('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB(alpha=.05)),
], metric, name, self.test_dir, self.test_dir)
labels = cf_cv.predict(['bad', 'good'], proba=True)
self.assertTrue(labels.shape == (2, 2))
self.assertEqual(type(labels), pd.DataFrame)
self.assertEqual(len(labels['pos'].values), 2)
self.assertEqual(len(labels['neg'].values), 2)
self.assertEqual(type(labels['neg'].values[0]), np.float64)
def test_calc_evaluation_report(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'MultinomialNB'
metric = 'f1'
cf_cv.train_save_metrics([('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB(alpha=.05))], metric,
name, self.test_dir, self.test_dir)
filename = os.path.join(self.test_dir, '_eval_report')
cf_cv.calc_evaluation_report(self.df_test['text'],
self.df_test['class'],
savefile=filename)
self.assertTrue(os.path.isfile(filename + "_" + name + ".csv"))
self.assertTrue(os.path.isfile(filename + "_" + name + "_average.csv"))
def test_unpickle(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'MultinomialNB'
metric = 'f1'
cf_cv.train_save_metrics([('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB(alpha=.05))], metric,
name, self.test_dir, self.test_dir)
savefile = os.path.join(self.test_dir, 'clf_cv.cv')
cf_cv.pickle(savefile)
new_cf_cv = ClassifierCv.unpickle(savefile)
texts = ['dont know that', 'nice good and bad']
predicted_labels_orig = cf_cv.predict(texts, proba=True)
predicted_labesl_new = new_cf_cv.predict(texts, proba=True)
self.assertTrue(all(predicted_labels_orig == predicted_labesl_new))
def test_plot_confusion_matrix_eval_data_normalize(self):
cf_cv = ClassifierCv(self.labels, self.texts)
name = 'MultinomialNB'
metric = 'f1'
cf_cv.train_save_metrics([
('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('clf', MultinomialNB(alpha=.05)),
], metric, name, self.test_dir, self.test_dir)
filename = os.path.join(self.test_dir,
name + 'eval_confusion_matrix.png')
cf_cv.labels_eval_real = ['pos', 'neg', 'neg', 'neg']
cf_cv.labels_eval_predicted = ['pos', 'neg', 'neg', 'neg']
cf_cv.plot_confusion_matrix(savefile=filename,
normalize=True,
use_evaluation_data=True)
self.assertTrue(os.path.isfile(filename))