数据竞赛（四）模型及调优

一、模型

1. GBDT

GBoost = GradientBoostingRegressor(n_estimators=3000, learning_rate=0.04,
                                   max_depth=6, max_features='sqrt',
                                   min_samples_leaf=15, min_samples_split=10, 
                                   loss='huber', random_state = 2333) 
GBoost.fit(train, target)
# 对训练集预测
pred_log1p = GBoost.predict(train)

2. XGB

model_xgb = xgb.XGBRegressor(colsample_bytree=0.4603, gamma=0.0468, 
                             learning_rate=0.05, max_depth=6, 
                             min_child_weight=1.7817, n_estimators=2200,
                             reg_alpha=0.4640, reg_lambda=0.8571,
                             subsample=0.5213, silent=1,
                             random_state =7, nthread = -1,

                            )
# y_train = target
test.columns = train.columns
# 进行模型训练
model_xgb.fit(train, target)
# 对训练集预测
pred_log1p = model_xgb.predict(train)

3. LGBM

folds = KFold(n_splits=5, shuffle=True, random_state=2333)

oof_lgb = np.zeros(len(train))
predictions_lgb = np.zeros(len(test))
feature_importance_df = pd.DataFrame()

for fold_, (trn_idx, val_idx) in enumerate(folds.split(train.values, target.values)):
    print("fold {}".format(fold_))
    trn_data = lgb.Dataset(train.iloc[trn_idx], label=target.iloc[trn_idx], categorical_feature=categorical_feats)
    val_data = lgb.Dataset(train.iloc[val_idx], label=target.iloc[val_idx], categorical_feature=categorical_feats)
#     trn_data = lgb.Dataset(train.iloc[trn_idx], label=target.iloc[trn_idx])
#     val_data = lgb.Dataset(train.iloc[val_idx], label=target.iloc[val_idx])

    num_round = 10000
    clf = lgb.train(params, trn_data, num_round, valid_sets=[trn_data, val_data], verbose_eval=500,
                    early_stopping_rounds=200)

    oof_lgb[val_idx] = clf.predict(train.iloc[val_idx], num_iteration=clf.best_iteration)
    OOF_lgb  = pd.DataFrame()
    OOF_lgb[fold_] = clf.predict(train, num_iteration=clf.best_iteration)

4. CatBoost

import catboost as cb
model_cb = cb.CatBoostRegressor(iterations=1000, depth=6, learning_rate=0.5,eval_metric = "R2",leaf_estimation_method = "Newton",
                            l2_leaf_reg=3
                            )
y_train = target
# 采用平滑后的y_train进行模型训练
model_cb.fit(train, y_train)
# 对训练集预测
pred_log1p = model_cb.predict(train)

5. ENet

ENet = make_pipeline(RobustScaler(), ElasticNet(alpha=1e-9, l1_ratio=.59, random_state=5))
# 采用平滑后的y_train进行模型训练
ENet.fit(train, log1p_y_train)
# 对训练集预测
pred_log1p = ENet.predict(train)

6. Lasso

lasso = make_pipeline(RobustScaler(), Lasso(alpha =1e-9, random_state=5))

# 采用平滑后的y_train进行模型训练
lasso.fit(train, log1p_y_train)
# 对训练集预测
pred_log1p = lasso.predict(train)

二、模型调优

1. hyperopt

（英文原版）https://districtdatalabs.silvrback.com/parameter-tuning-with-hyperopt

（中文译版）https://www.jianshu.com/p/35eed1567463

2. 网格搜索和随机搜索

model_selection.GridSearchCV

model_selection.RandomizedSearchCV