Unfreeze Tabular Models?

I am in the process of converting Rossmann to my own tabular datasets and I have a couple of questions:

  1. Do we unfreeze tabular models?
  2. If so, at what point do we unfreeze?

If you look at a tabular models architecture, we have nothing to freeze. There is only one module. That’s why if you follow the Rossmann, when we do a fit_one_cycle there is no slice() for the max_lr.

1 Like

Thank you.

How would this work? Pre-trained tabular models, on what?