AutoGluon BugReport for PR #1956

bug_report.py

from autogluon.tabular import TabularPredictor, TabularDataset


if __name__ == '__main__':
    path_prefix = 'https://autogluon.s3.amazonaws.com/datasets/CoverTypeMulticlassClassification/'
    path_train = path_prefix + 'train_data.csv'
    path_test = path_prefix + 'test_data.csv'

    label = 'Cover_Type'
    sample = 20000  # Number of rows to use to train
    train_data = TabularDataset(path_train)

    if sample is not None and (sample < len(train_data)):
        train_data = train_data.sample(n=sample, random_state=0).reset_index(drop=True)

    test_data = TabularDataset(path_test)

    fit_kwargs = dict(
        train_data=train_data,
        hyperparameters={
            'NN_TORCH': {},
            'GBM': {},
            'KNN': {},
        },
        presets='high_quality',
        time_limit=120,
        num_bag_folds=3,
        num_bag_sets=2,
    )

    predictor = TabularPredictor(
        label=label,
        eval_metric='log_loss',
    )

    predictor.fit(**fit_kwargs)
    predictor.leaderboard()
    predictor.persist_models('all')

    # works well
    we_model = predictor.fit_weighted_ensemble(base_models=['KNeighbors_BAG_L1', 'LightGBM_BAG_L1'], name_suffix='trial1')
    print(f'we_model={we_model}, expect ["WeightedEnsemble_L2_trial1"]')

    # not as expect
    we_model_refit = predictor.fit_weighted_ensemble(base_models=['KNeighbors_BAG_L1', 'LightGBM_BAG_L1'], name_suffix='trial2', refit_full=True)
    print(f'we_model_refit={we_model_refit}, expect ["WeightedEnsemble_L2_trial2_FULL"]')

    predictor.leaderboard()