search

tidymodels / recipes

Pipeable steps for feature engineering and data preprocessing to prepare for modeling
https://recipes.tidymodels.org
Other
550 stars 106 forks source link

Is there the autoencoder function in the tidymodels? #1292

Open JiahuaQu opened 3 months ago

JiahuaQu commented 3 months ago

We are very interested in using an autoencoder to reduce the dimension to input into a machine-learning model. We found this issue: https://github.com/tidymodels/recipes/issues/35. Did it mean that the autoencoder function has been embedded in tidymodels? How to use it in tidymodels? Thank you.

EmilHvitfeldt commented 3 months ago

Hello @JiahuaQu 👋

there is currently no support for autoencoders in tidymodels. That is not because we think they shouldn't be, but because we haven't spent enough time formalizing a good interface.

JiahuaQu commented 3 months ago

Thank you for answering me so quickly. Looking forward to your future support of autoencoder in tidymodels when available. If possible, could you recommend a way to connect an autoencoder process outside tidymodels with downstream analysis inside tidymodels?

EmilHvitfeldt commented 3 months ago

This project is not on any timelines right now, so no promises about when it will happen 😅

As for how to use with tidymodels. It will depend on where in the process you want to use it. The problem you are most likely to run into is related to data-leakage, as you are forced to train the autoencoder up front, and apply it to the data before splitting and having it used in the recipe.