def training_pipeline(text_cols, target_column, vectorizer_params, logistic_params):
return log_learner_time(
build_pipeline(
nlp_logistic_classification_learner(
text_feature_cols=text_cols,
target=target_column,
vectorizer_params=vectorizer_params,
logistic_params=logistic_params
)
), "tweet_sentiment_analysis")
def test_nlp_logistic_classification_learner():
df_train_binary = pd.DataFrame({
'id': ["id1", "id2", "id3", "id4"],
'x1': [10.0, 13.0, 10.0, 13.0],
"text1": ["banana manga", "manga açaí", "banana banana", "Manga."],
"text2": ["banana mamao", "manga açaí", "banana banana", "Manga."],
'y': [0, 1, 0, 1]
})
df_test_binary = pd.DataFrame({
'id': ["id4", "id4", "id5", "id6", "id5", "id6"],
'x1': [0.0, 3.0, 2.0, -13.0, 2.0, -13.0],
"text1": ["banana manga", "manga açaí", "banana banana", "Manga.", "banana manga", "manga açaí"],
"text2": ["banana manga", "manga açaí", "jaca banana", "Manga.", "jaca banana", "Manga."],
'y': [1, 0, 0, 1, 0, 1]
})
df_train_multinomial = pd.DataFrame({
'id': ["id1", "id2", "id3", "id4", "id3", "id4"],
'x1': [10.0, 13.0, 10.0, 13.0, 10.0, 13.0],
"text": ["banana manga", "manga açaí", "banana banana", "Manga.", "banana banana", "Manga."],
'y': [0, 1, 2, 0, 1, 2]
})
df_test_multinomial = pd.DataFrame({
'id': ["id4", "id4", "id5", "id6", "id5", "id6"],
'x1': [0.0, 3.0, 2.0, -13.0, 2.0, -13.0],
"text": ["abacaxi manga", "manga açaí", "banana banana", "Abacaxi.", "banana banana", "Manga."],
'y': [0, 1, 2, 0, 2, 1]
})
# test binomial case
learner_binary = nlp_logistic_classification_learner(text_feature_cols=["text1", "text2"],
target="y",
vectorizer_params={"min_df": 1},
logistic_params=None,
prediction_column="prediction")
predict_fn, pred_train, log = learner_binary(df_train_binary)
pred_test = predict_fn(df_test_binary)
expected_col_train = df_train_binary.columns.tolist() + ["prediction"]
expected_col_test = df_test_binary.columns.tolist() + ["prediction"]
assert Counter(expected_col_train) == Counter(pred_train.columns.tolist())
assert Counter(expected_col_test) == Counter(pred_test.columns.tolist())
assert pred_test.prediction.max() < 1
assert pred_test.prediction.min() > 0
assert (pred_test.columns == pred_train.columns).all()
# test multinomial case
learner_multinomial = nlp_logistic_classification_learner(text_feature_cols=["text"],
target="y",
vectorizer_params={"min_df": 1},
logistic_params={"multi_class": "multinomial",
"solver": "sag",
"max_iter": 200},
prediction_column="prediction")
predict_fn, pred_train, log = learner_multinomial(df_train_multinomial)
pred_test = predict_fn(df_test_multinomial)
expected_col_train = df_train_multinomial.columns.tolist() + ["prediction_0", "prediction_1", "prediction_2"]
expected_col_test = df_test_multinomial.columns.tolist() + ["prediction_0", "prediction_1", "prediction_2"]
assert Counter(expected_col_train) == Counter(pred_train.columns.tolist())
assert Counter(expected_col_test) == Counter(pred_test.columns.tolist())
assert (pred_test.columns == pred_train.columns).all()
# Time-based Train/Dev split
logging.info("Perform Train/Dev split")
train, dev = time_split_dataset(data,
time_column="input_time",
train_start_date="2017-01-01",
train_end_date="2018-01-01",
holdout_end_date="2019-01-01")
logging.info("Train set (Complaints 2017): {}".format(train.shape))
logging.info("Dev set (Complaints 2018): {}".format(dev.shape))
# Train
logging.info("Start baseline training")
baseline_p_fn, _, baseline_log = nlp_logistic_classification_learner(
train,
logistic_params={"solver": "liblinear"},
text_feature_cols=["text"],
target="target")
logging.info("Finished baseline training")
logging.info("Start model training")
p_fn, train_pred, log = nlp_logistic_classification_learner(
train,
vectorizer_params={
"strip_accents": "unicode",
"stop_words": "english"
},
logistic_params={
"solver": "liblinear",
"class_weight": "balanced"
},