0
$\begingroup$

I am working on a multilabel classification problem with both continuous and categorical features. For a single label problem, I might make use of a supervised encoder for my categorical features such as a TargetEncoder, CatBoost, etc. However, I don't know if these encoders can handle multilabel samples. Are there categorical encoders that are extensible to multilabel data? Also happy to look at any literature published on this specifically if people know of any.

Note: The categorical encoders in the scikit contrib category encoders package seem to be extensible to multilabel problems via the polynomial wrapper. However, I'm not sure if I'm interpreting the functionality correctly.

$\endgroup$
0
$\begingroup$

Something that I have used and worked well for me was binary encoding, which is a mix of OneHotEncoding and hash encoding. you can read more about it here:

https://towardsdatascience.com/all-about-categorical-variable-encoding-305f3361fd02

https://www.analyticsvidhya.com/blog/2020/08/types-of-categorical-data-encoding/

$\endgroup$
1
  • $\begingroup$ This looks helpful. I guess it avoids the multilabel issue because it is unsupervised, but also mitigates the memory and dimensionality issues from one-hot. Thanks! $\endgroup$
    – tensormoby
    Aug 20 at 15:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.