optunaを用いたlightGBMの実行時のエラー

Question

optunaのlightGBMを実行時にエラーが発生します．

以下の最後のlgb.LightGBMTunerCV実行時に
'set' object has no attribute 'startswith'
というエラーが発生してしまします．ここではcross validationを使用していますが，通常の学習でも同様のエラーが発生しました．

最初はlgb_trainの作成の仕方が悪いのかと疑いました．
しかし，optunaではなく本来のlightGBMモジュールを使用し同様に作成したデータを使用した際にはエラーは発生しませんでした．

どなたか理由のわかるかたお願いいたします．

import optuna.integration.lightgbm as lgb
from sklearn.model_selection import StratifiedKFold


lgb_train = lgb.Dataset(train_val_X, train_val_y)
lgb_test = lgb.Dataset(test_X, test_y, reference=lgb_train)

folds = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)


params = {'boosting_type': 'gbdt', 
          'objective': 'multiclass', 
          'metric': {'multi_logloss'},
          'num_class': 6 
         }

tuner_cv = lgb.LightGBMTunerCV(params, 
                              lgb_train,
                              early_stopping_rounds=20,
                              verbose_eval=True,
                              folds=folds
                              )

エラーを全文載せておきます．

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-72-9371d79f08d8> in <module>
     19                               early_stopping_rounds=20,
     20                               verbose_eval=True,
---> 21                               folds=folds
     22                               )

~/OneDrive/research/lab_env/lib/python3.7/site-packages/optuna/_experimental.py in wrapped_init(self, *args, **kwargs)
     84                     ),
     85                     ExperimentalWarning,
---> 86                     stacklevel=2,
     87                 )
     88 

~/OneDrive/research/lab_env/lib/python3.7/site-packages/optuna/integration/lightgbm_tuner/optimize.py in __init__(self, params, train_set, num_boost_round, folds, nfold, stratified, shuffle, fobj, feval, feature_name, categorical_feature, early_stopping_rounds, fpreproc, verbose_eval, show_stdv, seed, callbacks, time_budget, sample_size, study, optuna_callbacks, verbosity)

~/OneDrive/research/lab_env/lib/python3.7/site-packages/optuna/integration/lightgbm_tuner/optimize.py in __init__(self, params, train_set, num_boost_round, fobj, feval, feature_name, categorical_feature, early_stopping_rounds, verbose_eval, callbacks, time_budget, sample_size, study, optuna_callbacks, verbosity)

~/OneDrive/research/lab_env/lib/python3.7/site-packages/optuna/integration/lightgbm_tuner/optimize.py in higher_is_better(self)

AttributeError: 'set' object has no attribute 'startswith'

Answer 1

params = {'boosting_type': 'gbdt', 
          'objective': 'multiclass', 
          'metric': {'multi_logloss'},
          'num_class': 6 
         }

上記の{'multi_logloss'}の{}を削除してはどうでしょうか？

Comments

[kandaaa]

全く盲点でした.....
{}なくしたら直りました！！！ありがとうございます！！！！！