For Decoder:
We must then map the representation back to the multilabel categorical variables. This shall work equivalently with the AttrDecoder. Right now it is not very clear how to do this, but this article seems quite promising
https://www.learnopencv.com/multi-label-image-classification-with-pytorch/
Evaluation:
we must also be quite certain with the evaluation, respective loss function.
The above article discusses this issue as well.
In general, it should work in the same ways as the AttrEncoder and AttrDecoder. However, it must yield multi label predictions:
For Encoder: Learn for each categorical variable a seperate embedding. For this task the following link plus the AttrEncoder() implementation should be quite helpful: https://machinelearningmastery.com/how-to-prepare-categorical-data-for-deep-learning-in-python/
For Decoder: We must then map the representation back to the multilabel categorical variables. This shall work equivalently with the AttrDecoder. Right now it is not very clear how to do this, but this article seems quite promising https://www.learnopencv.com/multi-label-image-classification-with-pytorch/
Evaluation: we must also be quite certain with the evaluation, respective loss function. The above article discusses this issue as well.
Further Readings: 1.) https://users.ics.aalto.fi/jesse/talks/Multilabel-Part01.pdf 2.) https://medium.com/@thevatsalsaglani/training-and-deploying-a-multi-label-image-classifier-using-pytorch-flask-reactjs-and-firebase-c39c96f9c427