Deepcopy lightgbm model causing missing parameters?

[noklam]

Description

Deepcopy lightgbm model causing model lost parameters

Reproducible example

import pandas as pd
import numpy as np
import lightgbm as lgb
from copy import deepcopy

params = {
'objective': 'regression',
'verbose': -1,
'num_leaves': 3
}

X = np.random.rand(100,2)
Y = np.ravel(np.random.rand(100,1))
lgbm = lgb.train(params, lgb.Dataset(X,label=Y),num_boost_round=1)


print(lgbm.params)

## Deep copy will missing params
new_model = deepcopy(lgbm)
print(new_model.params)

Output:

{'objective': 'regression', 'verbose': -1, 'num_leaves': 3}
Finished loading model, total used 1 iterations
{}

Environment info

Windows 10, Python 3.7.5
LightGBM version or commit hash:
Tested with both lightgbm==2.3.1 and lightgbm==3.1.1 (latest pip version)

Command(s) you used to install LightGBM

Additional Comments

[noklam]

After digging into the source code, this seems to be the root cause, the __deepcopy__ only create the Boost object.

LightGBM/python-package/lightgbm/basic.py

Lines 2279 to 2282 in d6ebd06

	def __deepcopy__(self, _):
	model_str = self.model_to_string(num_iteration=-1)
	booster = Booster(model_str=model_str)
	return booster

    def __deepcopy__(self, _):
        model_str = self.model_to_string(num_iteration=-1)
        booster = Booster(model_str=model_str)
        return booster

Those commit are quite old already (2+years), I wonder why a custom implementation of __deepcopy__ is necessary in the first place.

[StrikerRUS]

Hi @noklam !
Indeed, params are not pre-saved at the Python wrapper layer after save/load process. However, underlying model is correct and it doesn't affect model behavior.
We already have the feature request for displaying original params used to train a model: #2613.

I wonder why a custom implementation of __deepcopy__ is necessary in the first place.

Custom __deepcopy__ is needed to make Booster class picklable.

[noklam]

If I understand correctly, the fix should be release in the next version and parameters could be saved with mode?

May I ask what's making it not picklable? Just trying to learn more about the issue

[StrikerRUS]

Unfortunately, there is no fix for that yet.

[noklam]

It seems picklable even without the deepcopy method? I try removed the __deepcopy__ and parameters is preserved and I can just pickle the model.

import lightgbm
from lightgbm import Booster
del Booster.__deepcopy__

params = {
'objective': 'regression',
'verbose': -1,
'num_leaves': 3
}

X = np.random.rand(100,2)
Y = np.ravel(np.random.rand(100,1))
lgbm = lgb.train(params, lgb.Dataset(X,label=Y),num_boost_round=1)


deepcopy_lgbm = deepcopy(lgbm)
lgbm.params, deepcopy_lgbm.params

[StrikerRUS]

Please refer to the following PR where custom __deepcopy__ was introduced and feel free to provide a PR if you think that this code can be removed.
#151

[StrikerRUS]

But I think that without custom __deepcopy__ cloned Booster will handle the same underlying cpp object as the original one.

[github-actions]

This issue 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.