Dataframe kindo-of-object in pytorch


This is a 100% open question, so, any idea would be more than welcome! I’m trying to emulate a similar structure to a pandas dataframe in pytorch (a data table, with named columns with different dtypes, with the possibility to run simple methods on them like sorting, etc), but after many days of searching, i haven’t been able to figure out if there is something similar or if can be achieved somehow with normal tensors.

I found something called torcharrow which seems to point in that direction, but the project is apparently abandoned and didn’t get past python 3.9.

Is there some pytorch object/project with this approach/functionality?

Thanks in advance!!

There is cuDF that is GPU powered and a drop-in-replace of pandas.
I guess that’s the most similar thing, but actually your goal is pretty hard :slight_smile:

Yeah, i know, but cudf (rapids) is also not yet available for python 3.12.

My work dev projects are basically in numpy+pandas (mostly), so, my goal was to see if i could basically replace both of them with pytorch.

I found that project i mentioned in my first post, torcharrow, which was literally that, pandas in pytorch, i was just hoping that there was another similar object/class in the pytorch ecosystem :slight_smile:

Thanks anyway!