lightgbm early_stopping_rounds parameter on fit method is deprecated.

[Redix8]

🐛 Bug

GBDTModel fit method has early_stopping_rounds parameter, but latest lightGBM (v4.0.0) has no parameter of it.
now it's using callbacks for early_stopping_rounds.

To Reproduce

Steps to reproduce the behavior:

import deepchem as dc
from lightgbm import LGBMRegressor

lgb = LGBMRegressor()
model = dc.models.GBDTModel(lgb)
model.fit(train_dataset)

File [.venv\lib\site-packages\deepchem\models\gbdt_models\gbdt_model.py:107, in GBDTModel.fit(self, dataset)
104 # Find optimal n_estimators based on original learning_rate and early_stopping_rounds
105 X_train, X_test, y_train, y_test = train_test_split(
106 X, y, test_size=0.2, random_state=seed, stratify=stratify)
--> 107 self.model.fit(
108 X_train,
109 y_train,
110 early_stopping_rounds=self.early_stopping_rounds,
111 eval_metric=self.eval_metric,
112 eval_set=[(X_test, y_test)])
114 # retrain model to whole data using best n_estimators * 1.25
115 if self.model.class.name.startswith('XGB'):

TypeError: LGBMRegressor.fit() got an unexpected keyword argument 'early_stopping_rounds'

Expected behavior

working as lightGBM(v3.3.5)

Environment

• OS: Windows
• Python version: 3.10
• DeepChem version: 2.7.1
• RDKit version (optional):
• TensorFlow version (optional):
• PyTorch version (optional):
• Any other relevant information:

Additional context

requirements on official document mentioned lightGBM version as "latest"
it need to be fixed to "v3.3.5" for current(v2.7.1 or before) version of deepchem
and maybe more implementation with callbacks for future version of deepchem

[rbharath]

Thanks for the report! I've marked this as a bug we should fix

[shreyasvinaya]

@rbharath can we close this issue, it is fixed now