jf

Hi, I am using CUDA 11.6, driver version 511.79 also on Windows 10. I build from source following the instructions in the documentation using CMake (added x64 generator) and VS2019.

Quick update, I cloned the repo again and build from source without CUBLAS and it works. If I enable the CUBLAS built in CMake the same error occurs.

This would be a really cool functionality to have! I'm considering the following scenario: I already built a DAG that gives me a baseline functionality or produced a delivarable I...

Here is a very rough sketch of what I had in mind: ```python # sample_module.py import pandas as pd from dev import mutate def my_function(input: pd.Series) -> pd.Series: return input...

re: `with_columns` -- I'll take a look tomorrow, I wasn't aware of it!

Nice, that makes a strong point for the use case that we shouldn't limit `@mutate` to impact a single function. 2. I would bind it the same way as pipe:...

Just documenting some thoughts while implementing to keep a record: 1. Using function pointers forces the `@mutate` functions to be placed after the definitions of the target functions --> nice...

Re Edge cases -- 1., 3. are ok! For 2.,4.,5. -- we could add the flag `allow_experimental_mode`. Then we have one version that offers basic, limited and safe functionality and...

Ok, this is very much WIP and my stab at this -- very open to just hearing what I did is ridiculous and I shouldn't complicate things so much :)...

Nice, glad it is going in the right direction. > @jernejfrank happy to look at the implenentation next, first took a look at the API. I really like it! >...