Optuna 파라미터 튜닝

 

import optuna

from optuna.samplers import TPESampler

 

def objective(trial):

  param = {

      "random_state":42,

      'learning_rate' : trial.suggest_loguniform('learning_rate', 0.01, 0.2),

      'bagging_fraction' :trial.suggest_float('bagging_fraction', 0.0, 1.0),

      "n_estimators":trial.suggest_int("n_estimators", 1, 1000),

      "max_depth":trial.suggest_int("max_depth", 1, 16),

      "num_leaves":trial.suggest_int("num_leaves",1, 100),

      "min_data_in_leaf": trial.suggest_int("=min_data_in_leaf", 2, 100),

      "feature_fraction": trial.suggest_float("feature_fraction", 0.0, 1.0),

      'lambda_l1': trial.suggest_float('lambda_l1', 0.0, 2.0),

      'lambda_l2': trial.suggest_float('lambda_l2', 0.0, 2.0),

      'path_smooth': trial.suggest_float('path_smooth', 0.0, 2.0),

  }

  X_t, X_v, y_t, y_v = train_test_split(train_X, train_Y, shuffle = True, test_size=0.2, random_state=42)

  

  lgbm = LGBMRegressor(**param)

  lgbm.fit(X_t, y_t,

          eval_set=[(X_t, y_t), (X_v,y_v)],

          early_stopping_rounds=35, verbose=100)

  lgbm_pred = lgbm.predict(X_v)

  RMSE = mean_squared_error(y_v, lgbm_pred, squared=False)

  return RMSE

 

sampler = TPESampler(seed=42)

study = optuna.create_study(

    study_name = 'lgbm_parameter_opt',

    direction = 'minimize',

    sampler = sampler,

)

study.optimize(objective, n_trials=10)

print("Best Score:",study.best_value)

print("Best trial",study.best_trial.params)