[FEA] Testing Feast in the multi-stage recommender pipeline for list type columns (multi-hot categorical)

[mkumari-ed]

🚀 Feature request

Currently materializing features using feast fails if there is a list type column i.e. multi-hot categorical feature in the dataset. We need some testing done around if feast supports such features or is it a MM issue and needs more deep dive engineering effort.

Motivation

This feature request is related to another GitHub issue NVIDIA-Merlin/Merlin#361

Your contribution

I am going to test on my end without using nvtabular for ETL if feast materialization goes through for such column in the dataset

[mkumari-ed]

[mkumari-ed]

[mkumari-ed]

Raw data looks like this

[mkumari-ed]

I have opened a FEA in systems repo for the same so closing it