Adapting Lesson 3 Notebook ColumnarModelData to Categorical Classification

Using a different data set but going off of the Rossman notebook.
This y values to predict are categorical(large cardinality) and so having issues getting that working with the fastai ColumnarModelData steps.
Any suggestions or anyone interested in working on the problem? Will share notebook and the data set is public.
Also, what is the closest pytorch model for this problem? Willing to use pure pytorch.

Additionally, getting this error which seems a problem with type in batch size but physically having trouble looking at things right now.
Here is the error:

TypeError: batch must contain numbers, dicts or lists; found <class ‘pandas.core.indexes.base.Index’>

Thanks in advance.

Hey gambit50

Have the same problem. Seems to be a little more complicated.
If you want to know more i think it is explained here:

Structured Learner

Still trying to wrap my head around it. Hope that helps.

I posted what you need to do to make it work here:

To achieve this I made a local copy of the and changed classes accordingly. This is clearly not the best way to do it longer term but works until I find the time to make the changes so that the classes can work with regression or classification without having to call different classes etc

If you want a pure Pytorch classification model then I can let you have a link to one but it won’t have Jeremy’s embedding of categorical variables. If does, however, show how to set it all up and manage it. I haven’t attached the link since my file is not sharable with public and I will have to sanitise it with some example data etc

1 Like