Kaggleで学ぶ機械学習（分類）-ハイパーパラメータチューニング-

# ファイルの読み込み・データ確認
application_train = pd.read_csv("../input/home-credit-default-risk/application_train.csv")
print(application_train.shape)
application_train.head()

import optuna

#学習用のデータセット作成
x_train = df_train.drop(columns=["TARGET", "SK_ID_CURR"])
y_train = df_train["TARGET"]
id_train = df_train[["SK_ID_CURR"]]

for col in x_train.columns:
    if x_train[col].dtype=="O":
        x_train[col] = x_train[col].astype("category")

#目的関数の定義
# 探索しないハイパーパラメータ
params_base = {
    "boosting_type": "gbdt",
    "objective": "binary",
    "metric": "auc",
    "verbosity": -1,
    "learning_rate": 0.05,
    "n_estimators": 100000,
    "bagging_freq": 1,
    "random_state": 123,
}

# 目的関数の定義
def objective(trial):
    # 探索するハイパーパラメータ
    params_tuning = {
        "num_leaves": trial.suggest_int("num_leaves", 8, 256),
        "min_child_samples": trial.suggest_int("min_child_samples", 5, 200),
        "min_sum_hessian_in_leaf": trial.suggest_float("min_sum_hessian_in_leaf", 1e-5, 1e-2, log=True),
        "feature_fraction": trial.suggest_float("feature_fraction", 0.5, 1.0),
        "bagging_fraction": trial.suggest_float("bagging_fraction", 0.5, 1.0),
        "lambda_l1": trial.suggest_float("lambda_l1", 1e-2, 1e+2, log=True),
        "lambda_l2": trial.suggest_float("lambda_l2", 1e-2, 1e+2, log=True),
    }
    params_tuning.update(params_base)
    
    # モデル学習・評価
    list_metrics = []
    cv = list(StratifiedKFold(n_splits=5, shuffle=True, random_state=123).split(x_train, y_train))
    list_fold = [0]  # 処理高速化のために1つめのfoldのみとする。
    for nfold in list_fold:
        idx_tr, idx_va = cv[nfold][0], cv[nfold][1]
        x_tr, y_tr = x_train.loc[idx_tr, :], y_train[idx_tr]
        x_va, y_va = x_train.loc[idx_va, :], y_train[idx_va]
        model = lgb.LGBMClassifier(**params_tuning)
        model.fit(x_tr,
                  y_tr,
                  eval_set=[(x_tr,y_tr), (x_va,y_va)],
                  early_stopping_rounds=100,
                  verbose=0,
                 )
        y_va_pred = model.predict_proba(x_va)[:,1]
        metric_va = roc_auc_score(y_va, y_va_pred) # 評価指標をAUCにする
        list_metrics.append(metric_va)
    
    # 評価指標の算出
    metrics = np.mean(list_metrics)
    
    return metrics

#最適化処理（探索の実行）
sampler = optuna.samplers.TPESampler(seed=123)
study = optuna.create_study(sampler=sampler, direction="maximize")
study.optimize(objective, n_trials=50, n_jobs=5)

#探索結果の確認
trial = study.best_trial
print("acc(best)={:.4f}".format(trial.value))
display(trial.params)