def getTrainedModel1(self):
# We build a matrix of LF votes for each comment ticket
LF_matrix = self.make_Ls_matrix(self.LF_set['comments'], self.LFs)
# Get true labels for LF set
Y_LF_set = np.array(self.LF_set['resolution'])
display(
lf_summary(sparse.csr_matrix(LF_matrix),
Y=Y_LF_set,
lf_names=self.LF_names.values()))
print("label coverage: " + label_coverage(LF_matrix))
mv = MajorityLabelVoter()
Y_train_majority_votes = mv.predict(LF_matrix)
print("classification report:\n" +
classification_report(Y_LF_set, Y_train_majority_votes))
Ls_train = self.make_Ls_matrix(self.train, self.LFs)
# You can tune the learning rate and class balance.
model = LabelModel(k=2, seed=123)
trainer = model.train_model(Ls_train,
n_epochs=2000,
print_every=1000,
lr=0.0001,
class_balance=np.array([0.2, 0.8]))
Y_train = model.predict(Ls_train) + Y_LF_set
print('Trained Label Model Metrics:')
scores = model.score((Ls_train[1], Y_train[1]),
metric=['accuracy', 'precision', 'recall', 'f1'])
print(scores)
return trainer, Y_train
def getTrainedModel2(self):
# Apply the LFs to the unlabeled training data
applier = PandasLFApplier(self.LFs)
L_train = applier.apply(self.train['comments'])
# Train the label model and compute the training labels
label_model = LabelModel(cardinality=2, verbose=True)
label_model.fit(L_train, n_epochs=500, log_freq=50, seed=123)
self.train['resolution'] = label_model.predict(
L=L_train, tie_break_policy="abstain")
df_train = self.train[self.train.resolution != self.ABSTAIN]
train_text = df_train.comments.tolist()
X_train = CountVectorizer(ngram_range=(1, 2)).fit_transform(train_text)
clf = LogisticRegression(solver="lbfgs")
clf.fit(X=X_train, y=df_train.resolution.values)
prob = clf.predict_proba(self.test)
if torch.cuda.is_available():
device = 'cuda'
else:
device = 'cpu'
end_model = EndModel([1000, 10, 2], seed=123, device=device)
end_model.train_model(
(self.train['comments'], self.test['comments']),
valid_data=(self.train['resolution'], self.test['comments']),
lr=0.01,
l2=0.01,
batch_size=256,
n_epochs=5,
checkpoint_metric='accuracy',
checkpoint_metric_mode='max')
return prob
candidate_dfs['dev'].curated_dsh,
model_type='curve',
figsize=(12, 7),
plot_title="Disease Associates Gene Dev PRC",
metric='PR',
font_size=16)
# In[21]:
label_model = LabelModel(k=2, seed=100)
label_model.train_model(validation_data[1][0],
n_epochs=1000,
verbose=False,
lr=0.01,
l2=2.067)
dev_predictions = convert_labels(label_model.predict(validation_data[1][1]),
'categorical', 'onezero')
dev_marginals = label_model.predict_proba(validation_data[1][1])[:, 0]
# In[22]:
plt.rcParams.update({'font.size': 16})
plt.figure(figsize=(10, 6))
plot_predictions_histogram(dev_predictions,
candidate_dfs['dev'].curated_dsh.astype(int).values,
title="Prediction Histogram for Dev Set")
# In[23]:
confusion_matrix(
convert_labels(candidate_dfs['dev'].curated_dsh.values, 'onezero',
log_train_every=50)
score = label_model.score((Ls[1], Ys[1]))
print('Trained Label Model Metrics:')
scores = label_model.score((Ls[1], Ys[1]),
metric=['accuracy', 'precision', 'recall', 'f1'])
mv = MajorityLabelVoter(seed=123)
print('Majority Label Voter Metrics:')
scores = mv.score((Ls[1], Ys[1]),
metric=['accuracy', 'precision', 'recall', 'f1'])
Y_train_ps = label_model.predict_proba(Ls[0])
Y_dev_p = label_model.predict(Ls[1])
"""
mv2 = MajorityClassVoter()
mv2.train_model(np.asarray(new_balance))
"""
#=np.asarray(new_balance))
#Y_baseline = mv2.predict(Ls[2])
pickling_on2 = open(
"data_encompassing/ar/ar_baseline_{}{}".format(flag0, flag), "wb")
pickle.dump(Y_baseline, pickling_on2)
print(Y_baseline)
# baseline majority:
"""