Karl Higley

See the cudf docs for [an explanation of how `NaN` works](https://docs.rapids.ai/api/cudf/stable/user_guide/Working-with-missing-data.html#Float-dtypes-and-missing-data). (Long story short, `NaN` is a float.) We _might_ be able to make the `NaN`s `s` instead, but I'm...

Which fields aren't supposed to be int32 and what error does it cause?

@jperez999 @bschifferer What's the next step with this?

@radekosmulski Thanks for the comments, I think you're entirely on target there! Re: PR etiquette etc, I'm really digging your thorough and well-thought out comments. The only thing I'd mention...

@radekosmulski I am also normally inclined to clean up my commits, but since all PRs get squashed before merging, I've mostly stopped. Piling commits on top of each other seems...

Waiting on rework of NVTabular notebooks

The only thing I know that serialized Python `lambdas` is `cloudpickle`, so I guess we could serialize the Workflow to `cloudpickle` and then deserialize it on the workers? Not too...

Yeah, good point @rjzamora. I would like to be able to do Dask computations across all the Merlin libraries, and also use Merlin graphs to run computations without Dask in...

@viswa-nvidia This PR was opened on the premise that we'd be working on offline batch recs generation in 22.08, as we'd planned before session-based bumped it out of the way....

I'm not able to reproduce these test failures locally, even in the `merlin_ci_runner` image. Going to try a re-run 🤷🏻