def test_make_single_prediction():
# Given
test_data = load_dataset(file_name='test.csv')
single_test_json = test_data[0:1]
# When
subject = make_prediction(input_data=single_test_json)
print(subject.get('predictions')[0])
# Then
assert subject is not None
assert isinstance((subject.get('predictions')[0]), np.integer)
assert math.ceil(subject.get('predictions')[0]) == 0
def test_make_multiple_predictions():
# Given
test_data = load_dataset(file_name='test.csv')
original_data_length = len(test_data)
multiple_test_json = test_data
# When
subject = make_prediction(input_data=multiple_test_json)
# Then
assert subject is not None
assert len(subject.get('predictions')) == 100
# We expect some rows to be filtered out
assert len(subject.get('predictions')) == original_data_length
def test_prediction_endpoint_returns_prediction(flask_test_client):
# Given
# Load the test data from the sainsbury_discontinued package
# This is important as it makes it harder for the test
# data versions to get confused by not spreading it
# across packages.
test_data = load_dataset(file_name=model_config.TESTING_DATA_FILE)
post_json = test_data[0:1].to_json(orient='records')
# When
response = flask_test_client.post('/v1/predict/classifier',
json=json.loads(post_json))
# Then
assert response.status_code == 200
response_json = json.loads(response.data)
prediction = response_json['predictions']
response_version = response_json['version']
assert prediction[0] == 0
assert response_version == _version
def test_prediction_endpoint_validation_200(flask_test_client):
# Given
# Load the test data from the regression_model package.
# This is important as it makes it harder for the test
# data versions to get confused by not spreading it
# across packages.
test_data = load_dataset(file_name=config.TESTING_DATA_FILE)
post_json = test_data.to_json(orient='records')
# When
response = flask_test_client.post('/v1/predict/classifier',
json=json.loads(post_json))
# Then
assert response.status_code == 200
response_json = json.loads(response.data)
# Check correct number of errors removed
assert len(response_json.get('predictions')) + len(
response_json.get('errors')) == len(test_data)
def run_training() -> None:
"""Train the model."""
# load training data
data = dm.load_dataset(file_name=config.TRAINING_DATA_FILE)
# train and test split
X_train, X_test, y_train, y_test = train_test_split(
data[config.FEATURES],
data[config.TARGET],
test_size=0.3,
random_state=1984,
stratify=data[config.TARGET])
# fit pipeline
pipeline.discontinued_pipe.fit(X_train[config.FEATURES], y_train)
# add model version to logs
_logger.info(f"saving model version: {_version}")
# save pipeline
dm.save_pipeline(pipeline_to_persist=pipeline.discontinued_pipe)
def make_predicitons(input_data):
_discontinued_identifier = dm.load_pipeline(file_name=config.PIPELINE_PATH)
result = _discontinued_identifier.predict(input_data)
prob = _discontinued_identifier.predict_proba(input_data)
prob_1 = prob[:, 1]
return result, prob_1
if __name__ == '__main__':
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
# load training data
data = dm.load_dataset(file_name=config.TRAINING_DATA_FILE)
# # train and test split
X_train, X_test, y_train, y_test = train_test_split(
data[config.FEATURES],
data[config.TARGET],
test_size=0.3,
random_state=1984,
stratify=data[config.TARGET])
# prediction and propensity
pred, prob_1 = make_predicitons(data[config.FEATURES])
data['pred'] = pred
data['prob_1'] = prob_1
# # determine classifcation report