def xgb_evaluate(
min_child_weight,
colsample_bytree,
max_depth,
subsample,
gamma,
reg_alpha,
reg_lambda,
):
target = 'visitors'
features = features_set_f0()
split = 0.33
seed = 177
full_data, ntrain, ntest = data_preparation()
trn = full_data[:ntrain]
x_train, x_valid, y_train, y_valid = train_test_split(trn[features].values,
trn[target].values,
test_size=split,
random_state=seed)
del full_data, trn
gc.collect()
xgb_params = dict()
xgb_params['objective'] = 'reg:linear'
xgb_params['eval_metric'] = 'rmse'
xgb_params['eta'] = 0.1
xgb_params['seed'] = seed
xgb_params['silent'] = True # does help
xgb_params['verbose_eval'] = False
xgb_params['nrounds'] = 500
xgb_params['max_depth'] = int(np.round(max_depth))
xgb_params['min_child_weight'] = int(np.round(min_child_weight))
xgb_params['colsample_bytree'] = colsample_bytree
xgb_params['subsample'] = subsample
xgb_params['gamma'] = gamma
xgb_params['alpha'] = reg_alpha
xgb_params['lambda'] = reg_lambda
xgb_clf = XgbWrapper(seed=seed, params=xgb_params)
xgb_clf.train(x_train, y_train, x_valid, y_valid)
return xgb_clf.best_score
def lgb_evaluate(min_child_sample, num_leaves, max_bin, min_child_weight,
subsample, subsample_freq, colsample_bytree, reg_alpha,
reg_lambda, feature_fraction, bagging_fraction):
target = 'visitors'
features = features_set_f0()
split = 0.33
seed = 177
full_data, ntrain, ntest = data_preparation()
trn = full_data[:ntrain]
x_train, x_valid, y_train, y_valid = train_test_split(trn[features].values,
trn[target].values,
test_size=split,
random_state=seed)
del full_data, trn
gc.collect()
lgb_params = dict()
lgb_params['objective'] = 'regression_l2'
lgb_params['metric'] = 'l2_root'
lgb_params['learning_rate'] = 0.1
lgb_params['random_state'] = seed
lgb_params['silent'] = True # does help
lgb_params['verbose_eval'] = False
lgb_params['n_estimators'] = 500
lgb_params['min_child_samples'] = int(np.round(min_child_sample))
lgb_params['num_leaves'] = int(np.round(num_leaves))
lgb_params['max_bin'] = int(np.round(max_bin))
lgb_params['subsample_freq'] = int(np.round(subsample_freq))
lgb_params['colsample_bytree'] = colsample_bytree
lgb_params['reg_alpha'] = reg_alpha
lgb_params['reg_lambda'] = reg_lambda
lgb_params['min_child_weight'] = min_child_weight
lgb_params['subsample'] = subsample
lgb_params['feature_fraction'] = feature_fraction
lgb_params['bagging_freq'] = int(np.round(bagging_fraction))
lgb_clf = LgbWrapper(params=lgb_params)
lgb_clf.train(x_train, y_train, x_valid, y_valid)
return lgb_clf.best_score
import sys, os
sys.path.append("../")
from general.preprocess import data_preparation
from general.ClfWrappers import XgbWrapper
from general.utilities import sub_to_csv
from features.f0 import features_set_f0
from cv.cv_02 import cross_validate
TARGET = 'visitors'
FEATURES = features_set_f0()
SEED = 177
print("Overfiting process initiating...")
xgb_params = dict()
xgb_params['objective'] = 'reg:linear'
xgb_params['eval_metric'] = 'rmse'
xgb_params['eta'] = 0.02
xgb_params['seed'] = SEED
xgb_params['silent'] = True # does help
xgb_params['verbose_eval'] = False
xgb_params['nrounds'] = 5000
xgb_params['early_stopping_rounds'] = 100
xgb_params['max_depth'] = 6
xgb_params['min_child_weight'] = 1
xgb_params['colsample_bytree'] = 0.724
xgb_params['subsample'] = 0.925
xgb_params['gamma'] = 0.512
xgb_params['alpha'] = 8.6
xgb_params['lambda'] = 1