def labeling_evaluation(df_train, df_test, label_model):
lfs = [
LabelingFunction.lf_ind_keyword, LabelingFunction.lf_short,
LabelingFunction.lf_cmp_re, LabelingFunction.lf_industry_keyword,
LabelingFunction.lf_surname_re, LabelingFunction.industry_cls
]
applier = PandasLFApplier(lfs=lfs)
L_train = applier.apply(df=df_train)
L_test = applier.apply(df=df_test)
Y_test = df_test.label.values
analysis = LFAnalysis(L=L_train, lfs=lfs).lf_summary()
if label_model == "majority":
majority_model = MajorityLabelVoter()
preds_train = majority_model.predict(L=L_train)
majority_acc = majority_model.score(
L=L_test, Y=Y_test, tie_break_policy="random")["accuracy"]
print(f"{'Majority Vote Accuracy:':<25} {majority_acc * 100:.1f}%")
df_train_filtered, preds_train_filtered = filter_unlabeled_dataframe(
X=df_train, y=preds_train, L=L_train)
return df_train_filtered, preds_train_filtered, analysis
if label_model == "weighted":
label_model = LabelModel(cardinality=len(
[c for c in dir(Polarity) if not c.startswith("__")]),
verbose=True)
label_model.fit(L_train=L_train, n_epochs=500, log_freq=100, seed=123)
probs_train = label_model.predict_proba(L_train)
label_model_acc = label_model.score(
L=L_test, Y=Y_test, tie_break_policy="random")["accuracy"]
print(f"{'Label Model Accuracy:':<25} {label_model_acc * 100:.1f}%")
df_train_filtered, probs_train_filtered = filter_unlabeled_dataframe(
X=df_train, y=probs_train, L=L_train)
preds_train_filtered = probs_to_preds(probs_train_filtered)
return df_train_filtered, probs_train_filtered, preds_train_filtered, analysis
def weak_supervisor(dataframe, model_type):
labeling_functions = [positive_labeling_function, positive1_labeling_function, negative_labeling_function,
negative1_labeling_function]
pandasApplier = PandasLFApplier(lfs=labeling_functions)
label_training_matrix = pandasApplier.apply(df=dataframe)
if model_type == "label_model":
# constructing a probabilistic label model
label_model = LabelModel(cardinality=2, verbose=True)
label_model.fit(L_train=label_training_matrix, n_epochs=300, log_freq=50, seed=123)
dataframe["weak_labels"] = label_model.predict(L=label_training_matrix)
print("dataframe shape: ", dataframe.shape)
dataframe = dataframe[dataframe["weak_labels"] != -1]
print("dataframe shape after filtering: ", dataframe.shape)
return dataframe
else:
majorityLabelVoter = MajorityLabelVoter()
dataframe["weak_labels"] = majorityLabelVoter.predict(L=label_training_matrix)
print("dataframe shape: ", dataframe.shape)
dataframe = dataframe[dataframe["weak_labels"] != -1]
print("dataframe shape after filtering: ", dataframe.shape)
return dataframe
]
# apply label functions
applier = PandasLFApplier(lfs=lfs)
# create a label matrix for the training set
L_train = applier.apply(df=data_train)
# create a label matrix for the test det
L_test = applier.apply(df=data_test)
# summary statistics for the LFs
lf_summary = LFAnalysis(L=L_train, lfs=lfs).lf_summary()
print(lf_summary)
# take the majority vote on a per-data point basis
majority_model = MajorityLabelVoter()
preds_train = majority_model.predict(L=L_train)
# use LabelModel to produce training labels
label_model = LabelModel(cardinality=2, verbose=True)
label_model.fit(L_train=L_train, n_epochs=500, log_freq=100, seed=123)
# result using majority-vote model
Y_test = data_test.label.values
majority_acc = majority_model.score(L=L_test,
Y=Y_test,
tie_break_policy="random")["accuracy"]
print(f"{'Majority Vote Accuracy:':<25} {majority_acc * 100:.1f}%")
# results using label model
label_model_acc = label_model.score(L=L_test,
Y=Y_test,