Save model on every iteration

[Qashqay]

Hello,

Is it possible to save intermediate model on every boosting iteration with the standard train function?
For example:

params  = {'num_leaves': 10, 'objective': 'rmse', 'save_freq': 1, 'save_path': '/tmp/'}
model = train(params, data)

and in /tmp/ we will have 10 files with identical to model.save_model('/tmp/model.txt') format.

Thank you

[jmoralez]

Hello @Qashqay, thank you for your interest in LightGBM. This isn't implemented in the library but since there are callbacks this is very easy to achieve if you implement your own callback. Every callback takes a CallbackEnv as argument, so you can use something like the following:

import lightgbm as lgb
import numpy as np

def save_model_callback(env):
    env.model.save_model(f'booster_{env.iteration}.txt')

ds = lgb.Dataset(np.random.rand(100, 2), np.random.rand(100))
model = lgb.train({'num_leaves': 10}, ds, num_boost_round=5, callbacks=[save_model_callback])

After running this you should see 5 files in your current directory (one for each iteration). You can modify the logic of that function to only save every x iterations or saving to a different path.

Please let us know if this helps.

[Qashqay]

Great, thank you. That's exactly what I wanted.

[jmoralez]

Great. I'm closing this issue but feel free to reopen if you run into some problem using this.

[gaebw]

running on windows 10:
Name: lightgbm
Version: 3.1.1
Summary: LightGBM Python Package
Home-page: https://github.com/microsoft/LightGBM
Author: None
Author-email: None
License: The MIT License (Microsoft)
Location: c:\users\info\anaconda3\lib\site-packages
Requires: scipy, scikit-learn, wheel, numpy

tried above code:
import lightgbm as lgb
import numpy as np

def save_model_callback(env):
env.model.save_model(f'booster_{env.iteration}.txt')

ds = lgb.Dataset(np.random.rand(100, 2), np.random.rand(100))
model = lgb.train({'num_leaves': 10}, ds, num_boost_round=5, callbacks=[save_model_callback])

Error:
AttributeError: function 'LGBM_DumpParamAliases' not found

[gaebw]

On ubuntu it works!

[jameslamb]

@gaebw LGBM_DumpParamAliases was introduced December 2, 2021 (#4829), so it's very surprising to see an error message about it given that you said you are using v3.1.1 (December 7, 2020).

I suspect that maybe you've built a newer version of the Python code (using python setup.py install --precompile) but linking to an older version of lib_lightgbm.so.

Please try building lightgbm's development version from the source code on GitHub, following the steps at https://github.com/microsoft/LightGBM/releases/tag/v3.3.1.

[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.