[python] Fix training on subset constructed without params

[StrikerRUS]

Fixed #5170.

Formally, we change the signature of the public get_params() method in this PR: it returns an empty dict instead of None. But I'm not sure about marking this PR a breaking change because previously this method could return empty dicts as well.

[jmoralez]

LGTM. After this change it's probably safe to change this function

LightGBM/python-package/lightgbm/basic.py

Lines 292 to 295 in 2871b53

	def param_dict_to_str(data):
	"""Convert Python dictionary to string, which is passed to C API."""
	if data is None or not data:
	return ""

to remove the case for None.

[jameslamb]

Looks good, thanks!

I left one very very minor suggestion, but marking "approve" so you can merge immediately after addressing it if you agree.

[StrikerRUS]

@jmoralez

After this change it's probably safe to change this function

Hmm, are you sure in this?
param_dict_to_str() is used to convert parameters in many places of the codebase. While in most cases before those lines we have something like this

LightGBM/python-package/lightgbm/basic.py

Line 2575 in 7d89ab4

params = {} if params is None else deepcopy(params)

there is no guarantee we haven't forgotten to do such None -> {} replacement somewhere. I think it's better to leave that explicit None handling in param_dict_to_str().

[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.