数据挖掘竞赛预测模型——GridSearch参数网格搜索

本案例使用lightGBM算法实现参数网格搜索

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.model_selection import GridSearchCV
import lightgbm as lgb

train_data = pd.read_csv('...')
train_label = pd.read_csv('...')
test_data = pd.read_csv('...')

train_x, test_x, train_y, test_y = train_test_split(train_data,train_label,test_size=0.3,random_state=2019)

cv_params = {   'n_estimators':[100,150,200,250,300],
                'num_leaves':[15,20,25,30,35,40,45,50],
                'max_depth':[3,4,5,6,7,8,9],
                'min_data_in_leaf':[18,19,20,21,22],
                'min_sum_hessian_in_leaf':[0.001,0.002],
                'feature_fraction':[0.6,0.7,0.8,0.9,1.0],
                'bagging_fraction':[0.6,0.7,0.8,0.9,1.0],
                'bagging_freq':[2,4,6,8,10],
                'lambda_l1':[1e-3,1e-2,0.0,0.1,0.2,0.3,0.4,0.5],
                'lambda_l2':[1e-3,1e-2,0.0,0.1,0.2,0.3,0.4,0.5],
                'learning_rate':[0.01,0.02,0.05,0.07,0.09,0.1,0.15,0.2]
                }
model = lgb.LGBMClassifier( 
                            boosting = 'gbdt',
                            objective = 'binary',  #分类用binary,多分类用multi-class,回归用regression
                            metrics = 'f1',
                            n_estimators = 100,
                            num_leaves = 30,     #搭配max_septh使用，取值<=2^(max_depth)，否则过拟合，单独调时可使得max_depth=-1，表示不限制树的深度
                            max_depth = 5,
                            min_data_in_leaf = 15,
                            min_sum_hession_in_leaf = 0.005,
                            feature_fraction = 0.8,
                            bagging_fraction = 0.8,
                            bagging_freq = 5,
                            lambda_l1 = 0.1,
                            lambda_l2 = 0.1,
                            learning_rate = 0.1
                            )
optimized_lgb = GridSearchCV(estimators=model, param_grid=cv_params, scoring='f1', cv=3, verbose=20, n_jobs=4)
optimized_lgb.fit(train_x, train_y)
test_pred = optimized_lgb.predicted(test_x)

best_model = optimized_lgb.best_estimator_
best_model.fit(train_x, train_y, eval_set=[(test_x, test_y)], eval_metric='f1', early_stopping_rounds=100)
print(best_model.feature_importances_)
best_params = optimized_lgb.best_params_
best_score = optimized_lgb.best_score_

与模型交叉验证搭配，先用gridsearch搜索模型最佳参数，再用交叉验证训练测试模型。

GridSearch参数网格搜索费时费力，往往能得到全局最佳参数；也可使用贝叶斯参数自动搜索，省时省力，但有时候容易得到的是局部最佳参数。可以先用贝叶斯自动搜索确定范围，再用GridSearch搜索确定最佳参数。