How to interpret/understand a tabular model?

Once a tabular model is predicting correctly, how do we interpret the factors the model the found?

Is there an api simliar to the collab learners’ weight.pca functionality?

1 Like

The pca functionality has been attached to all torch Tensors so for any torch Tensor you can use pca by simply calling pca() on it. For example if you have a tensor myTensor = torch.randn(40,40) you can use pca like myTensor.pca(). I can’t find it in the docs but the function is defined here.

1 Like

That’s good to know, thank you. I was disappointed when I searched the fastai docs for pca and found nothing!

I will take a look at the implementation of the collab leaner’s weight function to see if I can understand how to apply it to the tabular learner. I believe they have different embedding implementations, but under the hood they have to come down to a tensor of embedding weights which would have the pca function, correct?

1 Like

Sounds right! :slight_smile: