[R-package] prefer params to keyword argument in lgb.train()

[jameslamb]

From #4976 (comment).

Relates to #4913.

Today, when using lightgbm::lgb.train(), if you you pass different values to keyword argument obj and parameter "objective" in the params list, the value passed to the keyword argument will be preferred.

That is inconsistent with the way that the rest of the R package and the Python package works. Across LightGBM language-specific wrapper packages, it's expected that the values from params will take precedence over keyword arguments.

This PR proposes changing the R package to ensure that if objective or its aliases are passed through params, that value takes precedence over the keyword argument.

Why make this breaking change now?

Normally I'd propose that a change like this go through a deprecation cycle. However, it's been almost 4 months since the previous release (v3.3.1, not counting v3.3.2 since that was just a small patch to satisfy CRAN), and given the current state of open large PRs like #3234, #4630, and #4827 I expect it will be at least 2 months until the next, 4.0.0 release.

[jameslamb]

@shiyu1994 can you please help us diagnose why this license/cla check is stuck in pending?

#4985 and #5009 are also blocked by it. I believe the automation that checks whether contributors have signed the Contributor License Agreement is a Microsoft-wide configuration, not something specific to this repo, and I suspect something about it has changed.

[StrikerRUS]

Maybe you remember why is this line needed? Or more generally: why is lgb.check.obj() function needed?
Looks like there is a duplication with a similar check at cpp side. For example, there is no similar logic in the Python-package, but users can pass only registered objective names. Please consider the following example:

import numpy as np
import lightgbm as lgb

X = np.random.random((30, 5))
y = np.random.random((30))

lgb.LGBMRegressor(objective='no_such_objective').fit(X, y)

and the output:

---------------------------------------------------------------------------
LightGBMError                             Traceback (most recent call last)
<ipython-input-7-adb0d6a9439c> in <module>
----> 1 lgb.LGBMRegressor(objective='no_such_objective').fit(X, y)

D:\Miniconda3\lib\site-packages\lightgbm\sklearn.py in fit(self, X, y, sample_weight, init_score, eval_set, eval_names, eval_sample_weight, eval_init_score, eval_metric, early_stopping_rounds, verbose, feature_name, categorical_feature, callbacks, init_model)
    893             callbacks=None, init_model=None):
    894         """Docstring is inherited from the LGBMModel."""
--> 895         super().fit(X, y, sample_weight=sample_weight, init_score=init_score,
    896                     eval_set=eval_set, eval_names=eval_names, eval_sample_weight=eval_sample_weight,
    897                     eval_init_score=eval_init_score, eval_metric=eval_metric,

D:\Miniconda3\lib\site-packages\lightgbm\sklearn.py in fit(self, X, y, sample_weight, init_score, group, eval_set, eval_names, eval_sample_weight, eval_class_weight, eval_init_score, eval_group, eval_metric, early_stopping_rounds, verbose, feature_name, categorical_feature, callbacks, init_model)
    746         callbacks.append(record_evaluation(evals_result))
    747 
--> 748         self._Booster = train(
    749             params=params,
    750             train_set=train_set,

D:\Miniconda3\lib\site-packages\lightgbm\engine.py in train(params, train_set, num_boost_round, valid_sets, valid_names, fobj, feval, init_model, feature_name, categorical_feature, early_stopping_rounds, evals_result, verbose_eval, learning_rates, keep_training_booster, callbacks)
    269     # construct booster
    270     try:
--> 271         booster = Booster(params=params, train_set=train_set)
    272         if is_valid_contain_train:
    273             booster.set_train_data_name(train_data_name)

D:\Miniconda3\lib\site-packages\lightgbm\basic.py in __init__(self, params, train_set, model_file, model_str, silent)
   2608             params_str = param_dict_to_str(params)
   2609             self.handle = ctypes.c_void_p()
-> 2610             _safe_call(_LIB.LGBM_BoosterCreate(
   2611                 train_set.handle,
   2612                 c_str(params_str),

D:\Miniconda3\lib\site-packages\lightgbm\basic.py in _safe_call(ret)
    123     """
    124     if ret != 0:
--> 125         raise LightGBMError(_LIB.LGBM_GetLastError().decode('utf-8'))
    126 
    127 

LightGBMError: Unknown objective type name: no_such_objective

[jameslamb]

Looks to me like lgb.check.obj() has been a part of the R package since the very first PR introducing the R package, back in 2017.

https://github.com/microsoft/LightGBM/blame/9259a5318cd03951f2eefe378d9fb8db74457ee3/R-package/R/utils.R#L120

#168

I agree that in the current version of the R package, this isn't really necessary any more. And I'd support removing it, to remove the need to maintain a hard-coded list of objective names in the R package.

library(lightgbm)

data("EuStockMarkets")
lightgbm:::Booster$new(
    train_set = lightgbm::lgb.Dataset(
        data = EuStockMarkets[, c("SMI", "CAC", "FTSE")]
        , label = EuStockMarkets[, "DAX"]
    )
    , params = list(
        objective = "no_such_objective"
    )
)

Raises the following exception:

[LightGBM] [Fatal] Unknown objective type name: no_such_objective
Error in try({ : Unknown objective type name: no_such_objective
Error in initialize(...) : lgb.Booster: cannot create Booster handle

I'll open a separate PR tonight factoring it out.

[StrikerRUS]

Great, thanks!

[shiyu1994]

@shiyu1994 can you please help us diagnose why this license/cla check is stuck in pending?

#4985 and #5009 are also blocked by it. I believe the automation that checks whether contributors have signed the Contributor License Agreement is a Microsoft-wide configuration, not something specific to this repo, and I suspect something about it has changed.

Seems that has recovered now.

I think that would be related to the Microsoft CLA GitHub App.

[jameslamb]

Seems that has recovered now.

ah great, glad just closing and re-opening things fixed it. Thanks for checking @shiyu1994 and @StrikerRUS

[StrikerRUS]

LGTM, thank you!

[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.