Enable init_score for random forest boosting

[bili2002]

Summary

Currently, providing init_score to random forest training produces an error. After independent testing I have concluded that a fix is pretty straight forward.

Is there some hidden problem I am missing or should I submit a PR to comment on?

[jameslamb]

Please provide a minimum, reproducible example that shows evidence of the "bug". This link to a line of code is not enough evidence on its own.

[bili2002]

Here is an example:

import lightgbm as lgb
import numpy as np
import sklearn.datasets

X, y = sklearn.datasets.load_breast_cancer(return_X_y=True)

params = {
    "boosting_type": "rf",
    "objective": "binary",
    "bagging_freq": 1,
    "bagging_fraction": 0.5,
    "feature_fraction": 0.5,
    "num_leaves": 50,
    "metric": "binary_logloss",
    "verbose": -1,
}

print(
    lgb.train(
        params, lgb.Dataset(X, y, init_score=np.full(len(X), 0)), num_boost_round=50
    ).predict(X)
)

The code produces the following error:

[LightGBM] [Fatal] Check failed: (train_data->metadata().init_score()) == (nullptr) at /Users/runner/miniforge3/conda-bld/lightgbm_1726154828439/work/src/boosting/rf.hpp, line 48 .

Traceback (most recent call last):
  File "/Users/bilyanaindzheva/work/test.py", line 19, in <module>
    lgb.train(
  File "/Users/bilyanaindzheva/miniforge3/envs/test1/lib/python3.12/site-packages/lightgbm/engine.py", line 282, in train
    booster = Booster(params=params, train_set=train_set)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/bilyanaindzheva/miniforge3/envs/test1/lib/python3.12/site-packages/lightgbm/basic.py", line 3641, in __init__
    _safe_call(
  File "/Users/bilyanaindzheva/miniforge3/envs/test1/lib/python3.12/site-packages/lightgbm/basic.py", line 296, in _safe_call
    raise LightGBMError(_LIB.LGBM_GetLastError().decode("utf-8"))
lightgbm.basic.LightGBMError: Check failed: (train_data->metadata().init_score()) == (nullptr) at /Users/runner/miniforge3/conda-bld/lightgbm_1726154828439/work/src/boosting/rf.hpp, line 48 .

When boosting_type is gdbt the code executes correctly. Also, when init_score is not set then there is no error as well.