def test_run_train_test():
df_train, train_params = initial_processing(train_data, mode='train')
df_test, _ = initial_processing(test_data, mode='train')
tf = CatTransformer(train_params['cat_cols'])
tf.fit(df_train)
df_train_tf = tf.transform(df_train)
df_test_tf = tf.transform(df_test)
assert set(df_train_tf.columns.values) == set(df_test_tf.columns.values)
def run_train_test(ds_name, metric, params, obj):
path = _DATA_PATH + ds_name
with Profiler('initial feature selection'):
x_initial_raw, y_initial, _ = load_data(f'{path}/train.csv',
mode='train',
sample=_SAMPLE)
x_initial, ini_params = initial_processing(x_initial_raw, mode='train')
tf = CatTransformer(ini_params['cat_cols'])
# tf.fit(x_initial)
x_initial_tf = tf.fit_transform(x_initial)
selected_features, feat_list = ols_selection(x_initial_tf, y_initial,
obj)
hp_params = hyperopt_lgb(x_initial_tf[feat_list], y_initial, params,
obj)
print('selected features=', len(selected_features))
x_train_raw, y_train, _ = load_data(f'{path}/train.csv',
mode='train',
sample=_SAMPLE,
used_cols=selected_features)
x_test_raw, _, _ = load_data(f'{path}/test.csv', mode='test')
y_test = load_test_label(f'{path}/test-target.csv')
x_train, train_params = initial_processing(x_train_raw, mode='train')
x_test, test_params = initial_processing(x_test_raw, mode='test')
with Profiler('fit transform cat columns'):
x_test_rein = x_test.reindex(columns=train_params['used_cols'])
tf = CatTransformer(train_params['cat_cols'])
tf.fit(x_train)
x_train_tf = tf.transform(x_train)
x_test_tf = tf.transform(x_test_rein)
with Profiler('run train'):
model = lgb.train(hp_params, lgb.Dataset(x_train_tf, label=y_train),
600)
with Profiler('predict'):
y_train_out = model.predict(x_train_tf)
y_test_out = model.predict(x_test_tf)
train_err = metric(y_train, y_train_out)
test_err = metric(y_test, y_test_out)
return train_err, test_err
def test_cat_fit_test():
_, params = initial_processing(train_data, mode='train')
cat_cols = params['cat_cols']
tf = CatTransformer(cat_cols)
tf.fit(train_data[cat_cols])
res_df = tf.transform(test_data)
assert res_df['string_0'][0] == 2 / 5
assert res_df['string_0'][1] == 1 / 5
assert res_df['string_0'][2] == 2 / 5
assert np.isnan(res_df['string_0'][3])
assert np.isnan(res_df['string_0'][4])
assert np.isnan(res_df['string_1']).all()
def test_date_col_processing():
df, params = initial_processing(train_data, mode='train')
assert df['date_month_datetime_0'][0] == 1
assert df['date_month_datetime_0'][2] == 2
assert df['date_month_datetime_0'][3] == 11
assert df['date_weekday_datetime_0'][1] == 2
assert df['date_weekday_datetime_0'][3] == 0
assert df['date_weekday_datetime_0'][4] == 6
assert df['date_day_datetime_0'][1] == 3
assert df['date_day_datetime_0'][2] == 14
assert df['date_day_datetime_0'][4] == 30
"num_leaves": 200,
"feature_fraction": 0.70,
"bagging_fraction": 0.70,
'bagging_freq': 4,
"max_depth": -1,
"verbosity": -1,
"reg_alpha": 0.3,
"reg_lambda": 0.1,
"min_child_weight": 10,
'zero_as_missing': True,
'num_threads': 4,
'seed': 1
}
with Profiler('load data and perform feature selection'):
x_ini_raw, y_ini, _ = load_data(args.train_csv, sample=_SAMPLE)
x_initial, ini_params = initial_processing(x_ini_raw, mode='train')
tf = CatTransformer(ini_params['cat_cols'])
x_initial_tf = tf.fit_transform(x_initial)
selected_features, feat_list = ols_selection(x_initial_tf, y_ini, obj)
hp_params = hyperopt_lgb(x_initial_tf[feat_list], y_ini, params, obj)
print(f'{ len(selected_features)} features selected')
df_X_raw, df_y, _ = load_data(args.train_csv, used_cols=selected_features)
x_train, train_params = initial_processing(df_X_raw, mode='train')
with Profiler('fit transform cat columns'):
tf = CatTransformer(train_params['cat_cols'])
tf.fit(x_train)
x_train_tf = tf.transform(x_train)
with Profiler('run train'):
def test_cat_cols_frequency():
df, params = initial_processing(train_data, mode='train')
cat_cols = params['cat_cols']
assert set(cat_cols) == {'id_0', 'string_0', 'string_1'}
parser = argparse.ArgumentParser()
parser.add_argument('--test-csv', required=True)
parser.add_argument('--prediction-csv',
type=argparse.FileType('w'),
required=True)
parser.add_argument('--model-dir', required=True)
args = parser.parse_args()
start_time = time.time()
# load model
model_config_filename = os.path.join(args.model_dir, 'model_config.pkl')
with open(model_config_filename, 'rb') as fin:
model_config = pickle.load(fin)
model = model_config['model']
tf = model_config['cat_tf']
train_params = model_config['train_params']
x_test_raw, _, df = load_data(args.test_csv, mode='test')
x_test, test_params = initial_processing(x_test_raw, mode='test')
with Profiler('transform cat columns'):
x_test_rein = x_test.reindex(columns=train_params['used_cols'])
x_test_tf = tf.transform(x_test_rein)
df['prediction'] = model.predict(x_test_tf)
df[['line_id', 'prediction']].to_csv(args.prediction_csv, index=False)
print('Prediction time: {:0.2f}'.format(time.time() - start_time))