def making_data(data_path, target):
data = read_data(data_path)
X, y = extract_target(data, target)
cat_feats, num_feats = get_features_labels(X)
mu, sigma = 0, 0.1
noise = np.random.normal(mu, sigma, [X[num_feats].shape[0], X[num_feats].shape[1]])
num_feats_new = X[num_feats] + noise
data_fake = pd.concat([num_feats_new, X[cat_feats], y], axis=1)
return data_fake
def test_train_model(dataset_path: str, target_name: str, conf_path: str):
training_pipeline_params = read_training_pipeline_params(conf_path)
data = read_data(dataset_path)
X, y = extract_target(data, target_name)
X_transformed = full_transform(X)
X_train, X_test, y_train, y_test = split_train_val_data(
X_transformed, y, training_pipeline_params.splitting_params)
model = train_model(X_train, y_train,
training_pipeline_params.train_params)
assert isinstance(model, LogisticRegression)
def test_split_train_val_data(dataset_path: str, target_name: str,
conf_path: str):
training_pipeline_params = read_training_pipeline_params(conf_path)
data = read_data(dataset_path)
X, y = extract_target(data, target_name)
X_train, X_test, y_train, y_test = split_train_val_data(
X, y, training_pipeline_params.splitting_params)
assert len(X_train) > 0
assert len(X_test) > 0
assert len(y_train) > 0
assert len(y_test) > 0
def test_predict_model(dataset_path: str, target_name: str, conf_path: str):
training_pipeline_params = read_training_pipeline_params(conf_path)
data = read_data(dataset_path)
X, y = extract_target(data, target_name)
X_transformed = full_transform(X)
X_train, X_test, y_train, y_test = split_train_val_data(
X_transformed, y, training_pipeline_params.splitting_params)
model = train_model(X_train, y_train,
training_pipeline_params.train_params)
pred_labels, pred_proba = predict_model(model, X_test)
assert len(set(pred_labels)) == 2
assert max(pred_proba) < 1
def test_train_pipeline(dataset_path: str, target_name: str, conf_path: str):
training_pipeline_params = read_training_pipeline_params(conf_path)
data = read_data(dataset_path)
X, y = extract_target(data, target_name)
X_transformed = full_transform(X)
X_train, X_test, y_train, y_test = split_train_val_data(
X_transformed, y, training_pipeline_params.splitting_params)
model = train_model(X_train, y_train,
training_pipeline_params.train_params)
pred_labels, pred_proba = predict_model(model, X_test)
res = evaluate_model(y_test, pred_labels, pred_proba)
assert res['accuracy'] > 0
assert res['roc_auc_score'] > 0.5
def predict_pipeline_run(predict_pipeline_params):
logger.info(f"Start predict pipeline")
data = read_data(predict_pipeline_params.input_data_path)
# testing predict function on data without target
data, y = extract_target(data, 'target')
data_transformed = full_transform(data)
logger.info(f"Transformed data shape is {data_transformed.shape}")
model = load_model(predict_pipeline_params.dump_model)
pred_labels, pred_proba = predict_model(model, data_transformed)
pd.Series(pred_labels, index=data_transformed.index, name="prediction") \
.to_csv(predict_pipeline_params.result_path)
logger.info(f"Results written to directory")
def train_pipeline_run(training_pipeline_params):
logger.info(f"Start training pipeline")
data = read_data(training_pipeline_params.input_data_path)
X, y = extract_target(data, training_pipeline_params.target_name)
logger.info(f"X and y shape is {X.shape, y.shape}")
X_transformed = full_transform(X)
X_train, X_test, y_train, y_test = split_train_val_data(
X_transformed, y, training_pipeline_params.splitting_params)
model = train_model(X_train, y_train,
training_pipeline_params.train_params)
dump_model(training_pipeline_params.dump_model, model)
logger.info(f"model fitted and dumped")
pred_labels, pred_proba = predict_model(model, X_test)
res = evaluate_model(y_test, pred_labels, pred_proba)
logger.info(f"metrics is {res}")
def test_get_features_labels(dataset_path: str, target_name: str):
data = read_data(dataset_path)
X, y = extract_target(data, target_name)
cat_cols, num_cols = get_features_labels(X)
assert len(num_cols) > 0
def test_full_transform(dataset_path: str, target_name: str):
data = read_data(dataset_path)
X, y = extract_target(data, target_name)
X_transformed = full_transform(X)
assert max(X_transformed.describe().T['std']) < 2
assert max(abs(X_transformed.describe().T['mean'])) < 1
def test_extract_target(dataset_path: str, target_name: str):
data = read_data(dataset_path)
X, y = extract_target(data, target_name)
assert len(X) > 0
assert len(y) > 0