def run_training() -> None:
'''better train (implementing cv to avoid losing the data for validation'''
# divide dataset
data = load_dataset(file_name=config.TRAIN_DATA_FN)
X_train = data[config.TRAIN_FEATURES].sample(frac=1, random_state=42).drop(columns=config.TARGET)
y_train = data[config.TRAIN_FEATURES].sample(frac=1, random_state=42)[config.TARGET[0]]
pipeline.prep_pipeline.fit(X_train, y_train)
# prep input
prep_train = pipeline.prep_pipeline.transform(X_train)
# fit
np.random.seed(42)
pipeline.estimator_cv.fit(prep_train, y_train)
full_pipeline = Pipeline([
('prep_pipeline', pipeline.prep_pipeline),
('estimator', pipeline.estimator_cv.best_estimator_)
])
# print reulsts
print(f'best estimator: {pipeline.estimator_cv.best_estimator_}')
print(f'best estimator cv score: {pipeline.estimator_cv.best_score_}')
print(f'best estimator score: {pipeline.estimator_cv.best_estimator_.score(prep_train, y_train)}')
print(f'full pipe score: {full_pipeline.score(X_train, y_train)}')
_logger.info(f"save model version: {_version}")
save_pipeline(pipeline=full_pipeline)
def test_predict_proba():
file_name = 'test.csv'
data = load_dataset(file_name=file_name)
test_json = data[0:1].to_json(orient='records')
test_results = predict.make_prediction_proba(input_data=test_json)
assert test_results is not None
assert len(test_results['predictions'][0]) == 2
assert test_results['predictions'][0][0] <= 1
assert test_results['predictions'][0][0] >= 0
def test_predict():
file_name = 'test.csv'
data = load_dataset(file_name=file_name)
test_json = data[0:1].to_json(orient='records')
test_results = predict.make_prediction(input_data=test_json)
assert test_results is not None
assert isinstance(test_results['predictions'][0], int)
assert test_results['predictions'][0] <= 1
assert test_results['predictions'][0] >= 0
def test_prediction_endpoint(flask_test_client):
data = load_dataset(file_name=config.TEST_FILENAME)
post_json = data[0:1].to_json(orient='records')
_logger.info(f'POST JSON in test prediction endpoint: {post_json}')
_logger.info(
f'type of POST JSON in test prediction endpoint: {type(post_json)}')
response = flask_test_client.post('/v1/predict/lasso', json=post_json)
assert response.status_code == 200
response_json = json.loads(response.data)
response_pred = response_json.get('predictions')
response_version = response_json.get('version')
assert response_pred[0] <= 1 and response_pred[0] >= 0
assert response_version == model_version
def make_prediction(*, input_data: json) -> dict:
data = pd.read_json(input_data)
data = validate_data(data)
pred = pipeline.predict(data[config.FEATURES])
_logger.info(f"Making prediction with model version: {_version} "
f"Inputs: {data} "
f"Predictions: {pred}")
response = {'predictions': pred.tolist(), "version": _version}
return response
def make_prediction_proba(*, input_data) -> dict:
data = pd.read_json(input_data)
pred = pipeline.predict_proba(data[config.FEATURES])
response = {'predictions': pred.tolist(), "version": _version}
return response
if __name__ == '__main__':
data = load_dataset(file_name=config.TRAIN_DATA_FN)
X_train = data[config.TRAIN_FEATURES].sample(
frac=1, random_state=42).drop(columns=config.TARGET)
y_train = data[config.TRAIN_FEATURES].sample(
frac=1, random_state=42)[config.TARGET[0]]
# load model
print(pipeline.score(X_train, y_train))
import json
import pandas as pd
import numpy as np
from lasso.predict import make_prediction, make_prediction_proba
from lasso import config
from lasso.preprocess.data_management import load_dataset
from myapi import config as api_config
from myapi.logger_config import get_logger
from myapi.config import PREV_VER_PREDS_FILENAME
test_dataset = load_dataset(file_name=api_config.TEST_FILENAME)
_logger = get_logger(logger_name=__name__)
def capture_prev_ver_predictions():
df = test_dataset[111:444].copy()
test_json = df.to_json(orient='records')
test_preds = make_prediction_proba(input_data=test_json)
test_pred = pd.DataFrame(test_preds['predictions'])
test_pred_version = test_preds['version']
test_pred.to_csv(api_config.PACKAGE_ROOT /
api_config.PREV_VER_PREDS_FILENAME,
index=False)
_logger.debug(f'save previous version {test_pred_version} \
predictions: {api_config.PACKAGE_ROOT / api_config.PREV_VER_PREDS_FILENAME}'
)
import json
import pandas as pd
import numpy as np
import pytest
from tests.conftest import flask_test_client
from myapi.config import PACKAGE_ROOT, PREV_VER_PREDS_FILENAME, ACCETPTABLE_DIFF
from myapi.logger_config import get_logger
from lasso import config
from lasso.predict import make_prediction_proba
from lasso.preprocess.data_management import load_dataset
test_dataset = load_dataset(file_name=config.TEST_DATA_FN)
_logger = get_logger(logger_name=__name__)
@pytest.mark.differential
def test_differential(flask_test_client):
prev_ver_test_pred = pd.read_csv(PACKAGE_ROOT / PREV_VER_PREDS_FILENAME)
_logger.info('load previous version test predictions')
df = test_dataset[111:444].copy()
test_json = df.to_json(orient='records')
_logger.info(f'Test differential Inputs: {test_json}')
response = flask_test_client.post('/v1/predict_proba/lasso',
json = test_json)
pred_json = response.data
return validated_data
def validate_inputs(*, input_data: json):
data_schema = DataSchema(many=True)
input_dic = json.loads(input_data)
errors = None
try:
results = data_schema.load(input_dic)
except ValidationError as exc:
errors = exc.messages
# filter input
if errors:
validated_data = _filter_error_inputs(errors, input_dic)
validated_data = json.dumps(validated_data)
return validated_data, errors
return input_data, errors
if __name__ == '__main__':
data = load_dataset(file_name=config.TEST_FILENAME)
post_json = data[111:444].to_json(orient='records')
val_data, err = validate_inputs(input_data=post_json)
val_df = pd.read_json(val_data)
print(val_data)
