[Enhancement] CatBoost support in W&B

wandb --version && python --version && uname

• Weights and Biases version: wandb, version 0.8.31
• Python version: Python 3.7.2
• Operating System: MacOS Sierra

Description

I have successfully implemented W&B to capture parameters and save and upload my models to the cloud.

Although there is no callbacks support (via a named param) in Catboost.

If it’s trivial i’m happy to implement it for you if you can show me an example or point me to how you would this when no callbacks support is available.

Is there a custom WandBCallback implementation that can be adapted to work with Catboost?

What I Did

I looked on their GitHub repo and found this one conversation. And other such conversations as well. They talk about reading their debug logs or other log files written out during the training process.

[sekeratib]

Hi, is there any activity on this issue?

It would be great indeed, to have something like this working!

from wandb.catboost import WandbCallback

[Neronjust2017]

Hi, @neomatrix369 , I think I find the solution to use callback function to log metrics.

#1862

Hello! In CatBoost 0.26 we released support for callbacks, so your task is easily. We are preparing our new documentation, as this function still is not described, but you can look at example in our tests: https://github.com/catboost/catboost/blob/v0.26.1/catboost/python-package/ut/medium/test.py#L9726-L9738 I guess that’s a good sample to start with.