tune-sklearn VS spock

Of course not! :)

The Ray ecosystem is actually chalk full of integrations, from XGBoost Ray (https://docs.ray.io/en/master/xgboost-ray.html), to PyTorch on Ray (https://docs.ray.io/en/master/using-ray-with-pytorch.html), and of course hyperparameter search with Ray Tune for a variety of libraries, including Sklearn (https://github.com/ray-project/tune-sklearn).

If you mainly use scikit-learn, you should consider using tune-sklearn.

Just wanted to share a not widely known feature of PyCaret. By default, PyCaret's tune_model uses the tried and tested RandomizedSearchCV from scikit-learn. However, not everyone knows about the various advanced options tune_model() currently allows you to use such as cutting edge hyperparameter tuning techniques like Bayesian Optimization through libraries such as tune-sklearn, Hyperopt, and Optuna.

You might want to try out tune-sklearn as it seems like it works for catboost as well. I am trying it use tune-sklearn to speed up my scikit-learn hyperparameter tuning.

(Note: I wrote a library called spock that was originally based on dataclasses and then shifted to attrs. In the end attrs was just the better and more fully fledged library for what I needed so I’ve always preferred attrs over dataclasses since then)

Spock-Config allows one to create OO configuration files. That's how I roll. I currently use PYdantic settings and it's great. But it does not offer command-line re-configuration of what you have in the OO config file.

Check out more in the docs or on GitHub

We wrote Spock which actually sits in the middle ground between Hydra and OmegaConf (I’m of the same opinion that Hydra does a little too much feature wise). You can do hierarchical composition within the markdown of any JSON, YAML, or TOML files by simply using the config argument. No code needed to merge. Docs are here if you’re interested.

Spock