You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
If I understand correctly, there are two method in LightGBM for retraining: update and refit (+ train again with same params). Can we get support for both?
Personally, I think refit would be more important. First, train on huge vendor dataset. Then, refit with smaller client data. Although, I do not have any evidence refit is better than .update for this purpose, just my own thinking.
Also for huge model, being to pass some arguments directly to the booster would be awesome, So I can pass the model or the training and testing file or the file to retrain. Why? I noticed when I use the lightgbm CLI it uses may be 2TB memeory for 1TB data file. Where are in Python is just blow up, I didn't try on ML.net yet.
Great feature you added to be able to load a model from LightGBM and use it from ML.Net via c#
Triage: We see the value of the extra APIs but we currently do not have capacity to implement and support them. The current workaround is to do the refit/update outside of .NET and import the model afterwards.
Activity
torronen commentedon Jan 17, 2023
If I understand correctly, there are two method in LightGBM for retraining: update and refit (+ train again with same params). Can we get support for both?
https://lightgbm.readthedocs.io/en/latest/pythonapi/lightgbm.Booster.html#lightgbm.Booster.update
https://lightgbm.readthedocs.io/en/latest/pythonapi/lightgbm.Booster.html#lightgbm.Booster.refit
Personally, I think refit would be more important. First, train on huge vendor dataset. Then, refit with smaller client data. Although, I do not have any evidence refit is better than .update for this purpose, just my own thinking.
wil70 commentedon Jun 9, 2023
Hi @torronen and @luisquintanilla , this feature would be awesome. I would say both update and refit.
More detail here: https://stackoverflow.com/questions/73664093/lightgbm-train-vs-update-vs-refit#:~:text=Booster.refit%20%28%29%20does%20not%20change%20the%20structure%20of,add%20at%20most%201%20tree%20to%20the%20model.
Also for huge model, being to pass some arguments directly to the booster would be awesome, So I can pass the model or the training and testing file or the file to retrain. Why? I noticed when I use the lightgbm CLI it uses may be 2TB memeory for 1TB data file. Where are in Python is just blow up, I didn't try on ML.net yet.
Great feature you added to be able to load a model from LightGBM and use it from ML.Net via c#
TY
Wil
80LevelElf commentedon Aug 16, 2024
Hi!
Is there any updated on this feature? Really very needed)
matouskozak commentedon Oct 23, 2026
Triage: We see the value of the extra APIs but we currently do not have capacity to implement and support them. The current workaround is to do the refit/update outside of .NET and import the model afterwards.