fix: LightGBM 4.0+ compatibility for early_stopping_rounds=None

[Olcmyk]

Description

This PR fixes LightGBM 4.0+ compatibility issue where TypeError: early_stopping_round should be an integer. Got 'NoneType' is raised when early_stopping_rounds=None is passed to the model.

Starting from LightGBM 4.0, the lgb.early_stopping() function no longer accepts None as a parameter and requires an integer value. This PR modifies the code to only create the early stopping callback when early_stopping_rounds is not None.

Changes:

• Modified qlib/contrib/model/gbdt.py: Build callbacks list dynamically, only add early_stopping callback when rounds is not None
• Modified qlib/contrib/model/highfreq_gdbt_model.py: Apply same pattern for consistency and robustness

Motivation and Context

Related Issues:

• Fixes issue UnpicklingError: Forbidden class: qlib.contrib.data.handler.Alpha158 #2130 (partially - requires PR fix(unpickler): allow Alpha158/Alpha360 handlers and the standard dataset chain #2213 and PR Fix/pickle whitelist zscore #2230 first)
• Depends on PR Fix/pickle whitelist zscore #2230 (zscore pickle whitelist fix)

Problem:
After applying PR #2213 and PR #2230 (which fix the pickle whitelist issues), the DDG-DA workflow fails with:

TypeError: early_stopping_round should be an integer. Got 'NoneType'

Why this change is required:

1. LightGBM 4.0+ changed the API to require an integer for early_stopping_rounds
2. The DDG-DA workflow explicitly sets early_stopping_rounds=None to disable early stopping
3. The code directly passes None to lgb.early_stopping(), which is no longer allowed in LightGBM 4.0+

Root Cause:
In qlib/contrib/model/gbdt.py line 71-73, the code unconditionally creates an early stopping callback:

early_stopping_callback = lgb.early_stopping(
    self.early_stopping_rounds if early_stopping_rounds is None else early_stopping_rounds
)

When the value is None, LightGBM 4.0+ raises a TypeError.

Solution:
Only create the early stopping callback when the value is not None:

callbacks = []
early_stop_rounds = self.early_stopping_rounds if early_stopping_rounds is None else early_stopping_rounds
if early_stop_rounds is not None:
    callbacks.append(lgb.early_stopping(early_stop_rounds))

This pattern is already used in qlib/contrib/model/double_ensemble.py (lines 110-111).

How Has This Been Tested?

• Pass the test by running: pytest qlib/tests/test_all_pipeline.py under upper directory of qlib.
• If you are adding a new feature, test on your own test scripts.

Additional Testing:

• Tested with LightGBM 4.6.0 (latest version)
• Verified DDG-DA workflow proceeds past the LightGBM error
• Confirmed early stopping still works when a valid integer is provided
• Confirmed training works when early_stopping_rounds=None (no early stopping)

Test Environment:

• Python: 3.8.10
• OS: Linux (Ubuntu 22.04)
• LightGBM version: 4.6.0
• qlib version: 0.9.8.dev33

Test Command:

cd examples/benchmarks_dynamic/DDG-DA
rm -rf mlruns
python workflow.py run

Screenshots of Test Results (if appropriate):

1. Pipeline test: ✅ Passed
2. Your own tests:

Before the fix:

TypeError: early_stopping_round should be an integer. Got 'NoneType'
File "qlib/contrib/model/gbdt.py", line 71, in fit
    early_stopping_callback = lgb.early_stopping(...)

After the fix:

train tasks: 100%|████████████████████████████████████| 154/154 [05:31<00:00,  2.15s/it]
calc: 100%|█████████████████████████████████████████| 154/154 [00:01<00:00, 100.48it/s]

The LightGBM TypeError is fixed and training completes all 154 tasks successfully.

Note: After fixing this issue, the DDG-DA workflow encounters another bug (TypeError: unhashable type: 'slice'), which is a separate issue in the data selection module and will be addressed in a follow-up PR.

Types of changes

• Fix bugs
• Add new feature
• Update documentation

Additional Notes

Backward Compatibility:
This fix is fully backward compatible and works with both LightGBM < 4.0 and >= 4.0.

Dependencies:
This PR should be merged after:

• PR Fix/pickle whitelist zscore #2230 (zscore pickle whitelist)

Without these PRs, the zscore unpickling error will occur before reaching this LightGBM issue.

References:

• LightGBM 4.0 Release Notes: https://github.com/microsoft/LightGBM/releases/tag/v4.0.0
• LightGBM early_stopping API: https://lightgbm.readthedocs.io/en/latest/pythonapi/lightgbm.early_stopping.html

⚠️ Important: This PR depends on #2230

• This PR is based on PR Fix/pickle whitelist zscore #2230 and should be reviewed/merged after Fix/pickle whitelist zscore #2230 is merged.
• The commit history includes commits from Fix/pickle whitelist zscore #2230 because this bug only appears after that fix is applied.