【AI实战】xgb.XGBRegressor之多回归MultiOutputRegressor调参2（GPU训练模型）

xgb.XGBRegressor之多回归MultiOutputRegressor调参2（GPU训练模型）

• 环境

  • Ubuntu18.04
  • python3.6.9
  • TensorFlow 2.4.2
  • cuda 11.0
  • xgboost 1.5.2

• 依赖库

  import pandas as pd
  from sklearn.model_selection import train_test_split
  from sklearn.model_selection import GridSearchCV #网格搜索
  from sklearn.metrics import make_scorer
  from sklearn.metrics import r2_score
  from sklearn.ensemble import GradientBoostingRegressor
  from sklearn.multioutput import MultiOutputRegressor
  import xgboost as xgb
  import joblib
  

• 调参核心代码

  def tune_parameter(train_data_path, test_data_path, n_input, n_output, version):
      # 模型调参
      
      x,y = load_data(version, 'train', train_data_path, n_input, n_output)
      train_x,test_x,train_y,test_y = train_test_split(x,y,test_size=0.2,random_state=2022)
      
      gsc = GridSearchCV(
              estimator=xgb.XGBRegressor(seed=42,
                               tree_method='gpu_hist',
                               gpu_id=3),
              param_grid={"learning_rate": [0.05, 0.10, 0.15],
                          "n_estimators":[400, 500, 600, 700],
                          "max_depth": [ 3, 5, 7],
                          "min_child_weight": [ 1, 3, 5, 7],
                          "gamma":[ 0.0, 0.1, 0.2],
                          "colsample_bytree":[0.7, 0.8, 0.9],
                          "subsample":[0.7, 0.8, 0.9],
                          },
                          cv=3, scoring='neg_mean_squared_error', verbose=0, n_jobs=4)
  
      grid_result = MultiOutputRegressor(gsc).fit(train_x, train_y)
  
      #best_params = grid_result.estimators_[0].best_params_
      print('-'*20)
      print('best_params:')
      for i in range(len(grid_result.estimators_)):
          print(i, grid_result.estimators_[i].best_params_)
      
      model = grid_result
      
      pre_y = model.predict(test_x)
      print('-'*20)
      #计算决策系数r方
      r2 = performance_metric(test_y, pre_y)  
      print('test_r2 = ', r2)
  
  def performance_metric(y_true, y_predict):
      
      score = r2_score(y_true,y_predict)
      
      MSE=np.mean(( y_predict- y_true)**2)
      print('RMSE: ',MSE**0.5)
      MAE=np.mean(np.abs( y_predict- y_true))
      print('MAE: ',MAE)
      
      return score
      
  

  保存调参后的模型，增加下面代码即可

  joblib.dump(model, './ml_data/xgb_%d_%d_%s.model' %(n_input, n_output, version)) 
  

  调参：修改 param_grid 为自己的参数即可

  param_grid = {
  	"learning_rate": [0.05, 0.10, 0.15],
      "n_estimators":[400, 500, 600, 700],
      "max_depth": [ 3, 5, 7],
      "min_child_weight": [ 1, 3, 5, 7],
      "gamma":[ 0.0, 0.1, 0.2],
      "colsample_bytree":[0.7, 0.8, 0.9],
      "subsample":[0.7, 0.8, 0.9],
  }
  

  调参线程数量 n_jobs=4 ，可根据自己的机器设定，设置 n_jobs=-1 容易报错

  GPU参数设置：

  estimator=xgb.XGBRegressor(seed=42,
  	                             tree_method='gpu_hist',
  	                             gpu_id=3)
  

  tree_method=‘gpu_hist’ 表示使用GPU训练模型，
  gpu_id=3 表示设置第3块GPU，