Improving models via subsets

[@patrickleduc7712]

Thanks

[@AkaExcel]

Hi Vincent thank you for your video, please also do not forget to give code link

[@probabl_ai]

d0h! Added it in the shownotes too, it is here: github.com/probabl-ai/youtube-appendix/blob/main/13-sample-weights/sample-weights.ipynb

[@Amulya7]

Fascinating. Thanks!

[@probabl_ai]

You bet!

[@Amulya7]

Just a quick question:
Why was the sample_weight parameter in scikit-learn's models like Ridge designed to be passed during the fit method rather than being included as a model attribute or hyperparameter in the class definition? Wouldn't having sample_weight as part of the model initialization simplify its usage, particularly when integrating with pipelines?

[@probabl_ai]

You need the sample weights to be as long as the input data, which is only known at fit time.

[@TheMustafaKaan]

Would a different random seed change the ranking of the benchmark?

[@probabl_ai]

While I do not recall benchmarking for it, I am doing cross validation for the results. Is there a concern with the code?

[@TheMustafaKaan]

@@probabl_aiI supposed a single test set from the code. But convinced with cv since it reduce the variance.