User defined callbacks are not supported for GPU

[Enpen]

Problem:
I trained a model on GPU for a dataset. However,when i use clf.fit(X,y,callbacks=[mycallback1], i got an error as below:
<_catboost.CatBoostError: User defined loss functions, metrics and callbacks are not supported for GPU>. So how can i use my own callbacks when i train with GPU? Looking forward to your reply.

catboost version: 0.26

[Innixma]

Ran into this same issue with catboost v1.0

This makes writing code that works with both CPU and GPU very difficult, particularly when I am trying to guarantee that catboost returns within a given time limit (via callbacks).

[gaceladri]

Hello ✋ do you have any idea if callbacks would be supported in the future in catboost? thanks!

[Evgueni-Petrov-aka-espetrov]

@gaceladri ❤️
we hope to support user-defined metrics (high confidence) and loss-functions (less confidence) on gpu within few months from now
we thank you so much for using catboost 😻

[Innixma]

@Evgueni-Petrov-aka-espetrov That is amazing to hear!!

[lmsanch]

Any updates on this? It would be great to have pruning callbacks while training n GPU

[Evgueni-Petrov-aka-espetrov]

this is paused currently ☹️

[kgrushin]

Would be really nice to see this feature in the next versions of catboost

[EvanZ]

Any chance this is ever going to happen? What is the workaround?

[Evgueni-Petrov-aka-espetrov]

Any chance this is ever going to happen? What is the workaround?

a BS graduating this summer in SE has this a his thesis topic

formally speaking, workaround is using cpu

@EvanZ