models.append(("extra", model_extra))
scores = []
names = []
for name, model in models:
names.append(name)
scores.append(
np.sqrt(pp.score_model(model, train_X_reduced, train_y)).mean())
tab = pd.DataFrame({"Model": names, "Score": scores})
tab = tab.sort_values(by=['Score'], ascending=True)
print(tab)
for model in models:
model.fit(train_X, train_y)
averaged_models = em.AveragingModels(
models=[model_svr, model_ENet, model_KRR, model_xgb])
score_avg = np.sqrt(pp.score_model(averaged_models, train_X_reduced, train_y))
print(" Averaged base models score: {:.6f} ({:.6f})\n".format(
score_avg.mean(), score_avg.std()))
averaged_models.fit(train_X_reduced, train_y)
predicted_prices_averaged = np.expm1(averaged_models.predict(test_X_reduced))
print(predicted_prices_averaged)
my_submission = pd.DataFrame({
'Id': test.Id,
'SalePrice': predicted_prices_averaged
})
my_submission.to_csv('submission_avg.csv', index=False)
stacked_averaged_models = em.StackingAveragedModels(
random_seed = s
)
tree_models.append(("lgb_" + str(s), model_lgb))
#tree_models.append(("rf", model_rforest))
#tree_models.append(("xgb", model_xgb))
models.append(model_lgb)
tree_results = pp.get_cross_validate(tree_models, train_set_X, train_set_y,
folds = 10, seed = 2018, train_score = False, jobs = -1)
print(tree_results)
model_lgb.fit(train_set_X, train_set_y)
predicted = model_lgb.predict(test_set_X)
score_val = cvl.score_sq(test_set_y, predicted)
averaged_models = em.AveragingModels(models = models)
ensemble_models = []
ensemble_models.append(("averaged", averaged_models))
cross_val_table_avg = pp.get_validation_scores(ensemble_models, train_X_reduced, train_y, 5)
print(cross_val_table_avg)
pp.make_submission(averaged_models, train_X_reduced, train_y,
test_X_reduced, ids, filename = 'submission.csv')
lgbm_params = {
'task': 'train',
'boosting_type': 'gbdt',
'objective': 'regression',
linear_models.append(("enet", model_ENet))
linear_models.append(("enet", model_byr))
cross_val_table = get_validation_scores(tree_models, X_train, y_train, 5,
X_test, y_test)
print(cross_val_table)
cross_val_table = get_validation_scores(tree_models, train_X_reduced, train_y,
5)
print(cross_val_table)
cross_val_table_linear = get_validation_scores(linear_models, train_X_reduced,
train_y, 5)
print(cross_val_table_linear)
averaged_models = em.AveragingModels(
models=[model_lgb, model_byr, model_svr, model_ridge])
stacked_averaged_models = em.StackingAveragedModels(
base_models=[model_svr, model_lgb], meta_model=model_KRR)
averaged_plus = em.AveragingModels(
models=[averaged_models, model_GBoost, model_xgb], weights=[0.7, 0.2, 0.1])
averaged_plus_plus = em.AveragingModels(
models=[stacked_averaged_models, model_GBoost, model_xgb],
weights=[0.7, 0.2, 0.1])
ensemble_models = []
ensemble_models.append(("averaged", averaged_models))
ensemble_models.append(("stacked", stacked_averaged_models))
ensemble_models.append(("averaged_plus", averaged_plus))
ensemble_models.append(("averaged_plus_plus", averaged_plus_plus))
cross_val_table_ensemble = get_validation_scores(ensemble_models, X_train,
#early_stopping_rounds=100
verbose_eval=10,
random_seed=s)
tree_models.append(("lgb_" + str(s), model_lgb))
#tree_models.append(("rf", model_rforest))
#tree_models.append(("xgb", model_xgb))
cross_val_table = pp.get_validation_scores(tree_models, train_set_X,
train_set_y, 5)
print(cross_val_table)
model_lgb.fit(train_set_X, train_set_y)
predicted = model_lgb.predict(test_set_X)
score_val = cvl.score_sq(test_set_y, predicted)
averaged_models = em.AveragingModels(models=[model_lgb, model_rforest])
ensemble_models = []
ensemble_models.append(("averaged", averaged_models))
cross_val_table_avg = pp.get_validation_scores(ensemble_models,
train_X_reduced, train_y, 5)
print(cross_val_table_avg)
pp.make_submission(model_lgb,
train_X_reduced,
train_y,
test_X_reduced,
ids,
filename='submission.csv')
models.append(("lasso", model_lasso))
models.append(("ridge", model_ridge))
models.append(("svr", model_svr))
models.append(("ENet", model_ENet))
models.append(("KRR", model_KRR))
models.append(("byr", model_byr))
models.append(("rforest", model_rforest))
models.append(("xgb", model_xgb))
models.append(("GBoost", model_GBoost))
models.append(("lgb", model_lgb))
models.append(("lasso_lars", model_lasso_lars))
models.append(("lsvr", model_lsvr))
#models.append(("sgd", model_sgd))
#models.append(("extra", model_extra))
models.append(("average",
em.AveragingModels(models=[model_ridge, model_byr, model_xgb])))
######## cross validation #####################################################
def cross_validation_models(models_set, X_train_set, y_train_set):
scores = []
names = []
for name, model in models_set:
names.append(name)
scores.append(
np.sqrt(pp.score_model(model, X_train_set, y_train_set)).mean())
tab = pd.DataFrame({"Model": names, "Score": scores})
tab = tab.sort_values(by=['Score'], ascending=True)
return (tab)
train_y.ravel(),
folds=10,
seed=seed,
train_score=False,
jobs=2)
print(linear_results)
tree_results = pp.get_cross_validate(tree_models,
train_X_reduced,
train_y.ravel(),
folds=10,
seed=seed,
train_score=False)
print(tree_results)
averaged_models = em.AveragingModels(models=[model_lgb, model_KRR, model_svr])
stacked_averaged_models = em.StackingAveragedModels(
base_models=[model_KRR, model_lsvr, model_lgb], meta_model=model_ridge)
averaged_plus = em.AveragingModels(
models=[averaged_models, model_GBoost, model_xgb], weights=[0.7, 0.2, 0.1])
averaged_plus_plus = em.AveragingModels(
models=[stacked_averaged_models, model_GBoost, model_xgb],
weights=[0.7, 0.2, 0.1])
avg_full = em.AveragingModels(models=[
em.AveragingModels(models=[model_KRR, model_ridge, model_lsvr]),
em.AveragingModels(models=[model_lgb, model_GBoost, model_xgb])
])
ensemble_models = []
ensemble_models.append(("averaged", averaged_models))