[BUG] umap dirty_cat on colab

[maksim-mihtech]

With the last version of Pygraphistry I get the following error while running g.umap()

[lmeyerov]

Thanks @maksim-mihtech

Any chance you can share the file of df.sample(100).to_parquet('nodes.parquet') ?

cc @silkspace

[lmeyerov]

Also, what versions of python, dirty_cat , sci-kit, and pandas are you on?

[lmeyerov]

@maksim-mihtech it sounds like there is a repro in colab, can you share how you installed it for the repro, pip install graphistry[umap_learn] ?

[lmeyerov]

More info:

redteam-umap-gtc-gpu.ipynb, and I installed graphistry(CPU)/graphistryai

[maksim-mihtech]

@lmeyerov
It's installed like <pip install graphistry[ai]>

[maksim-mihtech]

[lmeyerov]

Thanks!

We are working on an end-run around this by upgrading from dirty-cat to skrub, you can track here: #626

[maksim-mihtech]

@lmeyerov

As you mentioned, I have tried the version(0.35.4+18.g60177c52) from dev/dev-skrub branch and it solved the problem. Thank you.

[lmeyerov]

That is great to hear!

We are still working on making that branch pass all our tests -- skrub is more picky on column names so we are adding more auto data cleaning to avoid skrub throwing exns on inputs that used to work -- but outside of that, the branch seems good!

[maksim-mihtech]

Thank you and whole Graphistry team!