How can I stop the log output of lightgbm?

Follow these points.

1. Use "verbose= False" in "fit" method.
2. Use "verbose= -100" when you call the classifier.
3. Keep "silent = True" (default).

Updated answer for 2024 (lightgbm>=4.3.0), describing how to suppress all log output from lightgbm (the Python package for LightGBM).

Answer

If using estimators from lightgbm.sklearn estimators:

• pass verbosity=-1 to estimator constructor

If using lightgbm.train(), lightgbm.cv(), lightgbm.Dataset():

• pass "verbosity": -1 through params keyword argument

Both interfaces emit a UserWarning if they encounter a conflict between keyword arguments and configuration passed other ways (**kwargs for scikit-learn, params dictionary for cv() / train(). To suppress those too:

• use warnings.filterwarnings("ignore", category=UserWarning)

Examples

Using Python 3.11, lightgbm==4.3.0, and scikit==1.4.1, with the following imports and setup for all examples:

import lightgbm as lgb
import warnings
from sklearn.datasets import make_regression

# create datasets
X, y = make_regression(
    n_samples=10_000,
    n_features=10,
)

This scikit-learn example with no explicit verbosity control…

model = lgb.LGBMRegressor(
    num_boost_round=10,
    num_leaves=31,
).fit(X, y)

…produces this output…

.../python3.11/site-packages/lightgbm/engine.py:172: UserWarning: Found `num_boost_round` in params. Will use it instead of argument
  _log_warning(f"Found `{alias}` in params. Will use it instead of argument")
[LightGBM] [Warning] num_iterations is set=10, num_boost_round=10 will be ignored. Current value: num_iterations=10
[LightGBM] [Info] Auto-choosing row-wise multi-threading, the overhead of testing was 0.000106 seconds.
You can set `force_row_wise=true` to remove the overhead.
And if memory is not enough, you can set `force_col_wise=true`.
[LightGBM] [Info] Total Bins 2550
[LightGBM] [Info] Number of data points in the train set: 10000, number of used features: 10
[LightGBM] [Info] Start training from score -1.188321

…but with verbosity controls…

warnings.filterwarnings("ignore", category=UserWarning)
model = lgb.LGBMRegressor(
    num_boost_round=10,
    num_leaves=31,
    verbosity=-1
).fit(X, y)

… does not produce any log output.

Similarly, using train() with no verbosity controls …

model = lgb.train(
    train_set=lgb.Dataset(X, label=y),
    params={
        "objective": "regression",
        "num_iterations": 10,
        "num_leaves": 31
    }
)

… produces this output …

.../python3.11/site-packages/lightgbm/engine.py:172: UserWarning: Found `num_iterations` in params. Will use it instead of argument
  _log_warning(f"Found `{alias}` in params. Will use it instead of argument")
[LightGBM] [Info] Auto-choosing row-wise multi-threading, the overhead of testing was 0.000136 seconds.
You can set `force_row_wise=true` to remove the overhead.
And if memory is not enough, you can set `force_col_wise=true`.
[LightGBM] [Info] Total Bins 2550
[LightGBM] [Info] Number of data points in the train set: 10000, number of used features: 10
[LightGBM] [Info] Start training from score -1.188321

…but with verbosity controls…

warnings.filterwarnings("ignore", category=UserWarning)
model = lgb.train(
    train_set=lgb.Dataset(X, label=y, params={"verbosity": -1}),
    params={
        "objective": "regression",
        "num_iterations": 10,
        "num_leaves": 31,
        "verbosity": -1
    }
)

… does not produce any log output.

In the list of parameters, which can be passed to LGBMClassifier or LGBMRegressor, there are no parameters responsible for logging – "verbose" or something like that. But using "*kwargs", this can be done – suppressing the output "verbosity(or verbose)=-1". The whole list of "*kwargs": https://lightgbm.readthedocs.io/en/latest/Parameters.html.

LGBMClassifier(verbosity=-1, random_state=42)