def main():
path_trial_folder_ulid = utl.get_path_trial_folder_ulid(c.CREATE_NEW)
print('Trial:', path_trial_folder_ulid)
# Mutable Paths
path_selected_features = path_trial_folder_ulid + c.SELECTED_FEATURES
path_log_tuneup = path_trial_folder_ulid + c.LOG_TUNEUP_FOR_SLC
path_model = path_trial_folder_ulid + c.MODEL_SLC
path_model_params = path_trial_folder_ulid + c.MODEL_PARAMS_SLC
path_roc_curve = path_trial_folder_ulid + c.ROC_CURVE_SLC
# Loading preprocessed data
train = joblib.load(c.PATH_TRAIN_PRP)
y = train[c.TARGET_COLUMN]
X = utl.except_for(train, c.TARGET_COLUMN)
if c.USE_SELECTED_FEATURES:
if not utl.exists(path_selected_features):
raise Exception("The file doesn't exist.")
selected_features = list(pd.read_csv(path_selected_features))
X = X[selected_features]
with utl.timer('tuneup'):
best_model = tuneup(models=c.MODEL_LIST,
params=c.MODEL_PARAMS,
X=X,
y=y,
scoring=c.SCORING,
direction=c.DIRECTION,
cv=c.CV,
n_splits=c.N_CV_SPLITS,
random_state=c.SEED,
n_trials=c.N_TRIALS_TUNE,
timeout=c.TIMEOUT,
n_jobs=-1,
path_model=path_model,
path_model_params=path_model_params,
path_log_tuneup=path_log_tuneup)
"""
Plot ROC curve
"""
with utl.timer('Plot ROC curve'):
viz.plot_roc_curve_with_cv(best_model,
X,
y,
cv=c.CV,
n_splits=c.N_CV_SPLITS,
test_size_ratio=c.TEST_SIZE_RATIO,
savepath=path_roc_curve)
def main():
# Loading data
train = pd.read_csv(c.PATH_TRAIN)
X = utl.except_for(train, c.TARGET_COLUMN)
X_test = pd.read_csv(c.PATH_TEST)
# Check if number of each class is inbalanced
y = train[c.TARGET_COLUMN]
print('class balance\n', y.value_counts())
# Create Profile of the data
if not utl.exists(c.PATH_PROFILE_REPORT_TRAIN):
with utl.timer('Create report of train'):
viz.create_profile(X, savepath=c.PATH_PROFILE_REPORT_TRAIN)
if not utl.exists(c.PATH_PROFILE_REPORT_TEST):
with utl.timer('Create report of test'):
viz.create_profile(
X_test, savepath=c.PATH_PROFILE_REPORT_TEST)
prp = Preprocessor()
prp.set_scaler(method=c.SCALING)
with utl.timer('Preprocessing train'):
train_prp = prp.exe(
df=train,
encoder=c.CAT_ENCODER,
exclusive_features=c.EXCLUSIVE_FEATURES,
dropped_features=c.DISCARDED_FEATURES,
thresh_nan_ratio_per_col=c.THRESH_NAN_RATIO_PER_COL,
thresh_corr=c.THRESH_CORR,
alt_num=c.ALT_NUM,
alt_cat=c.ALT_CAT,
path_train_prp=c.PATH_TRAIN_PRP)
y_prp = train_prp[c.TARGET_COLUMN]
X_prp = utl.except_for(train_prp, c.TARGET_COLUMN)
print('X_prp.shape:', X_prp.shape)
print('y_prp.shape:', y_prp.shape)
if not utl.exists(c.PATH_PROFILE_REPORT_TRAIN_PRP):
with utl.timer('Create report of train_prp'):
viz.create_profile(X_prp, savepath=c.PATH_PROFILE_REPORT_TRAIN_PRP)
with utl.timer('Preprocessing test'):
X_test_prp = prp.exe_test(
X=X_test,
y=None,
exclusive_features=c.EXCLUSIVE_FEATURES,
alt_num=c.ALT_NUM, alt_cat=c.ALT_CAT,
path_test_prp=c.PATH_TEST_PRP)
print('X_test_prp.shape:', X_test_prp.shape)
if not utl.exists(c.PATH_PROFILE_REPORT_TEST_PRP):
with utl.timer('Create report of test_prp'):
viz.create_profile(
X_test_prp, savepath=c.PATH_PROFILE_REPORT_TEST_PRP)
params=c.MODEL_PARAMS,
X=X,
y=y,
scoring=c.SCORING,
direction=c.DIRECTION,
cv=c.CV,
n_splits=c.N_CV_SPLITS,
random_state=c.SEED,
n_trials=c.N_TRIALS_TUNE,
timeout=c.TIMEOUT,
n_jobs=-1,
path_model=path_model,
path_model_params=path_model_params,
path_log_tuneup=path_log_tuneup)
"""
Plot ROC curve
"""
with utl.timer('Plot ROC curve'):
viz.plot_roc_curve_with_cv(best_model,
X,
y,
cv=c.CV,
n_splits=c.N_CV_SPLITS,
test_size_ratio=c.TEST_SIZE_RATIO,
savepath=path_roc_curve)
if __name__ == '__main__':
with utl.timer('Train'):
main()
X = utl.except_for(train, c.TARGET_COLUMN)
X_test = joblib.load(c.PATH_TEST_PRP)
if c.USE_SELECTED_FEATURES:
selected_features = list(pd.read_csv(path_selected_features))
X = X[selected_features]
X_test = X_test[selected_features]
"""
2. Predict
"""
model.fit(X, y)
y_pred = model.predict_proba(X_test) if c.USE_PREDICT_PROBA\
else model.predict(X_test)
df_prediction = pd.DataFrame({
c.TEST_ID: X_test[c.TEST_ID],
c.TARGET_COLUMN: y_pred
})
print('Prediction\n', df_prediction.head())
with open(path_submit, 'w', encoding='utf-8-sig') as f:
df_prediction.to_csv(f, index=False)
# !kaggle competitions submit -c titanic -f ML/Kaggle/Titanic/trial_models/01DXQSYFD4W2NX064GHV38N53Z/submit.csv -m "LogisticRegression"
if __name__ == '__main__':
with utl.timer('Test'):
main()
y_prp = train_prp[c.TARGET_COLUMN]
X_prp = utl.except_for(train_prp, c.TARGET_COLUMN)
print('X_prp.shape:', X_prp.shape)
print('y_prp.shape:', y_prp.shape)
if not utl.exists(c.PATH_PROFILE_REPORT_TRAIN_PRP):
with utl.timer('Create report of train_prp'):
viz.create_profile(X_prp, savepath=c.PATH_PROFILE_REPORT_TRAIN_PRP)
with utl.timer('Preprocessing test'):
X_test_prp = prp.exe_test(
X=X_test,
y=None,
exclusive_features=c.EXCLUSIVE_FEATURES,
alt_num=c.ALT_NUM, alt_cat=c.ALT_CAT,
path_test_prp=c.PATH_TEST_PRP)
print('X_test_prp.shape:', X_test_prp.shape)
if not utl.exists(c.PATH_PROFILE_REPORT_TEST_PRP):
with utl.timer('Create report of test_prp'):
viz.create_profile(
X_test_prp, savepath=c.PATH_PROFILE_REPORT_TEST_PRP)
if __name__ == '__main__':
with utl.timer('Preprocess'):
main()
path_trial_folder_ulid = utl.get_path_trial_folder_ulid(c.CREATE_NEW)
print('Trial:', path_trial_folder_ulid)
# Mutable Paths
path_selected_features = path_trial_folder_ulid + c.SELECTED_FEATURES
# Loading preprocessed data
train = joblib.load(c.PATH_TRAIN_PRP)
y = train[c.TARGET_COLUMN]
X = utl.except_for(train, c.TARGET_COLUMN)
if utl.exists_dir(c.PATH_LOG_FOLDER):
utl.mkdir(c.PATH_LOG_FOLDER)
model = ModelFactory(name=c.MODEL_NAME, params=c.MODEL_PARAMS).model
selected_features\
= select_features_by_rfe(
model=model, X=X, y=y,
ratio_max_n_features=c.RATIO_MAX_N_FEATURES,
path_selected_features=path_selected_features,
path_study_name_opt_features=c.STUDY_NAME_OPT_FIEATURES,
path_optuna_storage_opt_features=c.PATH_OPTUNA_STORAGE_OPT_FIEATURES)
print(selected_features)
if __name__ == '__main__':
with utl.timer('Select feature'):
main()