现代言情【机器学习】Optuna 机器学习模型调参（LightGBM 、XGBoost ）
⽂章⽬录
1. optuna 简介
在Kaggle⽐赛的过程中我发现了⼀个问题（⼤家的Kernel模型中包含了众多c超参数设置，但是这些参数是如何设置的呢？），并在
Discussion中提出了我的问题，并得到了众多⼤佬的回答，如下：
关于回答我汇总后发现都提到了关于optuna库的使⽤，optuna是什么呢？optuna是⼀个使⽤python编写
的超参数调节框架。⼀个极简的optuna 的优化程序中只有三个最核⼼的概念，⽬标函数(objective)，单次试验(trial)，和研究(study). 其中 objective 负责定义待优化函数并指定参/超参数数范围，trial 对应着 objective 的单次执⾏，⽽ study 则负责管理优化，决定优化的⽅式，总试验的次数、试验结果的记录等功能。
下⾯举⼀个简单的栗⼦，有助于⼤家的理解：
定义，求取得最⼤值时，的取值？
import optuna
def objective (trial ):
x = trial .suggest_uniform ('x', -10, 10)
y = trial .suggest_uniform ('y', -10, 10)
return (x + y ) ** 2
study = optuna .create_study (direction ='maximize')
study .optimize (objective , n_trials =100)
print (study .best_params )
print (study .best_value )
2. LGBM 和XGBoost 调参汇总
2.1 LGBM
2.1.1 定义Objective
x ,y ∈(−10,10)f (x )=(x +y )2x ,y
from lightgbm import LGBMRegressor
import optuna
ics import mean_squared_error
del_lection import train_test_split, KFold
import optuna.integration.lightgbm as oplgb
def objective(trial):
X_train, X_test, y_train, y_test=train_test_split(data, target, train_size=0.3)# 数据集划分
param ={
'metric':'rm',
'random_state':48,
'n_estimators':20000,
'reg_alpha': trial.suggest_loguniform('reg_alpha',1e-3,10.0),
'reg_lambda': trial.suggest_loguniform('reg_lambda',1e-3,10.0),
'colsample_bytree': trial.suggest_categorical('colsample_bytree',[0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]), 'subsample': trial.suggest_categorical('subsample',[0.4,0.5,0.6,0.7,0.8,1.0]),
'learning_rate': trial.suggest_categorical('learning_rate',[0.006,0.008,0.01,0.014,0.017,0.02]), 'max_depth': trial.suggest_categorical('max_depth',[5,7,9,11,13,15,17,20,50]),
'num_leaves': trial.suggest_int('num_leaves',1,1000),
'min_child_samples': trial.suggest_int('min_child_samples',1,300),
开机花屏'cat_smooth': trial.suggest_int('cat_smooth',1,100)
}
lgb=LGBMRegressor(**param)
lgb.fit(X_train, y_train, eval_t=[(X_test, y_test)], early_stopping_rounds=100, verbo=Fal)
pred_lgb=lgb.predict(X_test)手绘技巧
rm = mean_squared_error(y_test, pred_lgb, squared=Fal)
return rm
熨烫2.1.2 调参try
ate_study(direction='minimize')
嘴唇黑是什么原因n_trials=50# try50次
study.optimize(objective, n_trials=n_trials)
2.1.3 绘图
optuna.visualization.plot_optimization_history(study)# 绘制
optuna.visualization.plot_parallel_coordinate(study)#
optuna.visualization.plot_param_importances(study)#
2.1.4 最佳参数
params=study.best_params
头尖额窄面相params['metric']='rm'
2.2 XGBOOST
2.2.1 定义Objectove
def objective(trial):
data = train.iloc[:,:-1]
target = train.target
train_x, test_x, train_y, test_y = train_test_split(data, target, test_size=0.3, random_state=42)
param ={
'lambda': trial.suggest_loguniform('lambda',1e-3,10.0),
'alpha': trial.suggest_loguniform('alpha',1e-3,10.0),
'colsample_bytree': trial.suggest_categorical('colsample_bytree',[0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0]), 'subsample': trial.suggest_categorical('subsample',[0.4,0.5,0.6,0.7,0.8,1.0]),
'learning_rate': trial.suggest_categorical('learning_rate',
[0.008,0.009,0.01,0.012,0.014,0.016,0.018,0.02]),
'n_estimators':4000,
'max_depth': trial.suggest_categorical('max_depth',[5,7,9,11,13,15,17,20]),
'random_state': trial.suggest_categorical('random_state',[24,48,2020]),
'min_child_weight': trial.suggest_int('min_child_weight',1,300),
}
model = xgb.XGBRegressor(**param)
model.fit(train_x, train_y, eval_t=[(test_x, test_y)], early_stopping_rounds=100, verbo=Fal)
preds = model.predict(test_x)
rm = mean_squared_error(test_y, preds, squared=Fal)
return rm
2.2.2 调参try
study = ate_study(direction='minimize')
n_trials=1
study.optimize(objective, n_trials=n_trials)
print('Number of finished trials:',ials))
print("------------------------------------------------")
print('Best trial:', study.best_trial.params)
print("------------------------------------------------")
ials_dataframe())
print("------------------------------------------------")
2.2.3 绘图
optuna.visualization.plot_optimization_history(study).show()
猪肝瘦肉汤#plot_parallel_coordinate: interactively visualizes the hyperparameters and scores
optuna.visualization.plot_parallel_coordinate(study).show()
'''plot_slice: shows the evolution of the arch. You can e where in the hyperparameter space your arch went and which parts of the space were explored more.'''
optuna.visualization.plot_slice(study).show()
optuna.visualization.plot_contour(study, params=['alpha',
#'max_depth',
'lambda',
'subsample',
'learning_rate',
鸡蛋不能和什么一起吃'subsample']).show()
#Visualize parameter importances.
optuna.visualization.plot_param_importances(study).show()
#Visualize empirical distribution function
optuna.visualization.plot_edf(study).show()
2.2.4 最佳参数
params=study.best_params
