Skip to content

[python-package] Latest version of LGBM taking over 17 hours to train, 3.3.5 version takes 5 minutes #6932

Open
@ign3

Description

@ign3

Hi, as said in the title, I'm trying to update the lightgbm version to the latest, 4.6.0, as requested by the security department of my company.

At the step that is taking much longer, I'm using default parameters, so I didn't change the code at all:

estimator = lgb.LGBMRegressor(n_jobs=-1, verbosity=2)

Then, to ensure the parameters were the same as the default parameters of the 3.3.5 version, I forced them:

params = {'boosting_type': 'gbdt', 'class_weight': None, 'colsample_bytree': 1.0, 'importance_type': 'split', 'learning_rate': 0.1, 'max_depth': -1, 'min_child_samples': 20, 'min_child_weight': 0.001, 'min_split_gain': 0.0, 'n_estimators': 100, 'n_jobs': -1, 'num_leaves': 31, 'objective': None, 'random_state': None, 'reg_alpha': 0.0, 'reg_lambda': 0.0, 'silent': 'warn', 'subsample': 1.0, 'subsample_for_bin': 200000, 'subsample_freq': 0, 'verbosity': 2}
estimator = lgb.LGBMRegressor(**params)

The model in the 4.6.0 version is slowly iterating and using more features, but as I said, the time taken is absurd. I guess the difference lays in one of the default parameters the 4.6.0 has that I haven't changed with the parameters I passed, but I'm not sure.

I installed lightgbm through pip.

Does anyone have an idea what it might be? Thanks

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions