[SOLVED]
Problem 1) TabularDataBunch is using DataFrame index in place of data.x
(where one would expect a matrix of the actual data.
Solution: Upgrade to ‘1.0.34’
Problem 2) Training was not working, and I was getting “NaNs” in place of loss (see screenshot)
to float
Solution: convert all numerical columns to float32
. It was that easy
The original post below
TabularDataBunch is using DataFrame index in place of data.x
(where one would expect a matrix of the actual data.
I am using standard procs variable: procs = [FillMissing, Categorify, Normalize]
As a consequence, learning is not working.
See screenshot below for clarification.
I am using version 1.0.24