Question on model evaluation

[VRM1]

Thanks for the implementation! I was curious as to why you have both mask and mask_bc under miscelanea.py -> function def test_mie_ll as dataset[:][2]. Shouldn't the mask_bc = dataset[:][1] ? i.e., we need to be applying the mask when encoding and then generate the data. dataset[:][2] refers to the variable nan_mask (in file datasets.py) which is essentially all ones. It will be great if you can clarify this.

Thanks

[adrianjav]

Hi @VRM1. You are completely right, we should feed the observed mask dataset[:][1] to the model, and then use the missing mask to compute the missing imputation error and likelihood.

This is fixed in my follow-up repository for another project, but somehow I forgot to update this repository.

Thanks a lot for noticing, I will update it now.