def test_make_single_prediction():
# Given
test_data = load_dataset(file_name='test.csv')
single_test_input = test_data[0:1]
# When
subject = make_prediction(input_data=single_test_input)
# Then
assert subject is not None
assert isinstance(subject.get('predictions')[0], float)
assert math.ceil(subject.get('predictions')[0]) == 112476
def test_make_multiple_predictions():
# Given
test_data = load_dataset(file_name='test.csv')
original_data_length = len(test_data)
multiple_test_input = test_data
# When
subject = make_prediction(input_data=multiple_test_input)
# Then
assert subject is not None
assert len(subject.get('predictions')) == 1451
# We expect some rows to be filtered out
assert len(subject.get('predictions')) != original_data_length
def capture_predictions() -> None:
"""Save the test data predictions to a CSV."""
save_file = 'test_data_predictions.csv'
test_data = load_dataset(file_name='test.csv')
# we take a slice with no input validation issues
multiple_test_input = test_data[99:600]
predictions = make_prediction(input_data=multiple_test_input)
# save predictions for the test dataset
predictions_df = pd.DataFrame(predictions)
# hack here to save the file to the regression model
# package of the repo, not the installed package
predictions_df.to_csv(f'{config.PACKAGE_ROOT}/{save_file}')
def test_model_prediction_differential(
*, save_file: str = 'test_data_predictions.csv'):
"""
This test compares the prediction result similarity of
the current model with the previous model's results.
"""
# Given
# Load the saved previous model predictions
previous_model_df = pd.read_csv(f'{config.PACKAGE_ROOT}/{save_file}')
previous_model_predictions = previous_model_df.predictions.values
test_data = load_dataset(file_name=model_config.TESTING_DATA_FILE)
multiple_test_input = test_data[99:600]
# When
current_result = make_prediction(input_data=multiple_test_input)
current_model_predictions = current_result.get('predictions')
# Then
# diff the current model vs. the old model
assert len(previous_model_predictions) == len(current_model_predictions)
# Perform the differential test
for previous_value, current_value in zip(previous_model_predictions,
current_model_predictions):
# convert numpy float64 to Python float.
previous_value = previous_value.item()
current_value = current_value.item()
# rel_tol is the relative tolerance – it is the maximum allowed
# difference between a and b, relative to the larger absolute
# value of a or b. For example, to set a tolerance of 5%, pass
# rel_tol=0.05.
assert math.isclose(previous_value,
current_value,
rel_tol=model_config.ACCEPTABLE_MODEL_DIFFERENCE)
def predict():
if request.method == 'POST':
# Step 1: Extract POST data from request body as JSON
json_data = request.get_json()
_logger.debug(f'Inputs: {json_data}')
# Step 2: Validate the input using marshmallow schema
input_data, errors = validate_inputs(input_data=json_data)
# Step 3: Model prediction
result = make_prediction(input_data=input_data)
_logger.debug(f'Outputs: {result}')
# Step 4: Convert numpy ndarray to list
predictions = result.get('predictions').tolist()
version = result.get('version')
# Step 5: Return the response as JSON
return jsonify({
'predictions': predictions,
'version': version,
'errors': errors
})