[BUG] record_stats=False not working as expected

Describe the bug record_stats=False corrupts validation dataset. However, it works fine, when I set record_stats=True on the validation dataset. Steps/Code to reproduce bug

Workflow Processor:

• Train dataset record_stats=True and Validation dataset record_stats=False -> as you can see CumCount Max is 896 instead of 299.

• Train dataset record_stats=True and Validation dataset record_stats=True -> as you can see CumCount Max is now 299.

Expected behavior I am expecting this output when Validation dataset record_stats is set to False.

Environment details (please complete the following information):

• Environment location: [Bare-metal, Docker, Cloud(specify cloud provider)]
• Method of NVTabular install: [conda, Docker, or from source]
  • If method of install is [Docker], provide docker pull & docker run commands used

Installed 0.2 version from here…https://pypi.org/project/nvtabular/

And using it with Rapids 0.16 as I need pivot().

Additional context Add any other context about the problem here.

[rjzamora]

Perhaps it would be reasonable to add a Workflow parameter to specify a list of columns that should pass through NVTabular unchanged - Thoughts on this @benfred ? I’m honestly unsure how often a feature like this would be used.

[rjzamora]

is there a better way to pass down id column untouched via NVTabular

Rather than allowing the Categorify to act on all categorical columns (the default), you can specify a subset with Categorify(columns=<your-list>).