Closed hengzhe-zhang closed 10 months ago
Hello! Sorry for the slow reply, and thanks for your work! I like that skorch is used. I should add that hyperparameters for specific models should be tuned, if possible. While we have the default configuration for FT-Transformer, we do not have it for other models (though, good defaults for them may exist as well!).
FYI, the rtdl
package is now replaced with two packages:
rtdl_num_embeddings
(announcement, link)rtdl_revisiting_models
(link)The new packages only implement the papers, so scikit-learn interface is not planned for them. I have been keeping this issue open for quite some time to make it more visible, I hope it helped. Thank you for your repository!
Hello! I have written a scikit-learn interface for the RTDL package (https://github.com/hengzhe-zhang/scikit-rtdl). I rely on the skorch to avoid coding errors, and set the default parameters based on the parameters presented in your paper. Hoping you will like it!