def retrain():
global pipeline
predictors_name, target_name = my_libarary.getPredTargetNamesName(
model_name, client_url, db_name, collection_name)
if (pipeline == None):
loadPipeline()
#get data as json
data = request.get_json()
predictors_name, target_name = my_libarary.getPredTargetNamesName(
model_name, client_url, db_name, collection_name)
#separate features and target
features, target = my_libarary.preprocess(data, predictors_name,
target_name)
#encode target
target_enc = my_libarary.encode_target(target)
#load json to a pandas dataframe
features_df = pd.DataFrame(features)
#train the module
pipeline.fit(features_df, target_enc)
#save the model on database
my_libarary.updateModel(pipeline, model_name, client_url, db_name,
collection_name)
#prepare the response
return jsonify({'status': 'the model is fited'})
def predict():
predictors_name, target_name = my_libarary.getPredTargetNamesName(
model_name, client_url, db_name, collection_name)
global pipeline
if (pipeline == None):
loadPipeline()
#get data as json
dictList = request.get_json()
#projection
projected = my_libarary.projection(dictList, predictors_name)
#load json to a pandas dataframe
features_df = pd.DataFrame(projected)
#preprocessing data and get the prediction(1 or 0)
predictions = pipeline.predict(features_df).tolist()
#decode predictions
predictions_decd = my_libarary.decode_target(predictions)
#prepare the response
response = []
for point, pred in zip(projected, predictions_decd):
point['Attrition_pred'] = pred
response.append(point)
return jsonify(response)
def test_module():
global pipeline
predictors_name, target_name = my_libarary.getPredTargetNamesName(
model_name, client_url, db_name, collection_name)
if (pipeline == None):
loadPipeline()
#get data as json
json = request.get_json()
data = json['dataset']
n_splits = json['n_splits']
stratified = json['stratified']
shuffle = json['shuffle']
n_jobs = json['n_jobs']
predictors_name, target_name = my_libarary.getPredTargetNamesName(
model_name, client_url, db_name, collection_name)
#separate features and target
features, target = my_libarary.preprocess(data, predictors_name,
target_name)
#load json to a pandas dataframe
features_df = pd.DataFrame(features)
#encode target
target_enc = my_libarary.encode_target(target)
#cross_validate function and multiple metric evaluation
cv = 5 #default value
if (stratified):
cv = StratifiedKFold(n_splits=n_splits, shuffle=shuffle)
elif (shuffle):
cv = ShuffleSplit(n_splits=n_splits)
else:
cv = n_splits
scoring = ['f1', 'recall', 'precision', 'roc_auc']
cv_score = cross_validate(pipeline,
features_df,
target_enc,
n_jobs=n_jobs,
cv=cv,
return_train_score=True,
scoring=scoring)
auc = np.mean(cv_score['test_roc_auc'].tolist())
f1 = np.mean(cv_score['test_f1'].tolist())
precision = np.mean(cv_score['test_precision'].tolist())
recall = np.mean(cv_score['test_recall'].tolist())
train_time = np.mean(cv_score['test_recall'].tolist())
my_libarary.updateModelScores(model_name, model_name, client_url, db_name,
collection_name, f1, precision, recall, auc,
train_time)
return jsonify({
'score_roc_auc': auc,
'score_f1': f1,
'score_precision': precision,
'score_recall': recall,
'train_time': train_time
})