[python-package] add more hints in sklearn.py

[jameslamb]

Contributes to #3756.

Adds some more type hints to places in sklearn.py that are missing them.

Notes for Reviewers

I intentionally avoided hints on anything "array-like", since those will require some extra research and should be handled separately (#3756 (comment)).

[jmoralez]

I think this should be Sequence[int], given that for example range(5) is an iterable but is not a valid type for this.

[jameslamb]

🤩 Excellent point. I was just following the docstring and didn't consider this. Thanks very much for noting it!!!

I tried the following from the root of the repo, just to see what would happen:

sample code (click me)

from pathlib import Path
import numpy as np
import lightgbm as lgb
from sklearn.datasets import load_svmlight_file

rank_example_dir = Path('examples/lambdarank')
X_train, y_train = load_svmlight_file(str(rank_example_dir / 'rank.train'))
X_test, y_test = load_svmlight_file(str(rank_example_dir / 'rank.test'))
q_train = np.loadtxt(str(rank_example_dir / 'rank.train.query'))
q_test = np.loadtxt(str(rank_example_dir / 'rank.test.query'))
gbm = lgb.LGBMRanker(n_estimators=10)
gbm.fit(
    X_train,
    y_train,
    group=q_train,
    eval_set=[(X_test, y_test)],
    eval_group=[q_test],
    eval_at=range(3),
    callbacks=[
        lgb.early_stopping(10),
        lgb.reset_parameter(learning_rate=lambda x: max(0.01, 0.1 - 0.01 * x))
    ]
)

And you're right...passing a range for eval_at causes a failure when serializing the parameters to string to pass them through the C API functions.

  File ".../site-packages/lightgbm/basic.py", line 326, in param_dict_to_str
    raise TypeError(f'Unknown type of parameter:{key}, got:{type(val).__name__}')
TypeError: Unknown type of parameter:eval_at, got:range

Given that, I think the hint here should be even stricter than typing.Sequence. Since this keyword argument is passed directly through to params and there's no other code in LightGBM manipulating its value, I think it can only accept values that are valid for lightgbm.basic.param_dict_to_str().

For eval_at, I think that means only a list of ints or tuple of ints is valid. param_dict_to_str() supports list, tuple, and set, but set isn't appropriate for eval_at because sets aren't iterable (e.g. don't have any ordering).

LightGBM/python-package/lightgbm/basic.py

Line 323 in 3d4e08e

if isinstance(val, (list, tuple, set)) or is_numpy_1d_array(val):

I just pushed 81c234f which:

• sets the hint for eval_at to Union[List[int], Tuple[int]]
• replaces use of the word "iterable" in the relevant docstrings with "list or tuple of int"

[jameslamb]

@jmoralez I won't merge this until you have a chance to respond, since what I did here is slightly different than what you suggested.

[StrikerRUS]

LGTM, thanks!

[jmoralez]

Thanks!

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.