Error FailToDownloadPythonDependencies on mnist example

[mwbryant]

Ubuntu 20.04 LTS

On my machine when running the mnist example I get this message:

thread 'main' panicked at 'called Result::unwrap() on an Err value: FailToDownloadPythonDependencies("pillow, numpy | error: No such file or directory (os error 2)")', burn-dataset/src/source/huggingface/mnist.rs:40:14
note: run with RUST_BACKTRACE=1 environment variable to display a backtrace`

I tracked this down to https://github.com/burn-rs/burn/blob/0110ccc96471a404479c5e67801f22309bcf7682/burn-dataset/src/source/huggingface/downloader.rs#L196 which calls the command python when I believe it requires python3

The fix for me was to apt install python-is-python3

I then had problems with pandas requring a higher version of numpy. The fix for that was python -m pip install --upgrade numpy

I believe both these are not problems with the library code but with the installation instructions (of which there appear to be none) and might be worth mentioning when those are created.

[nathanielsimard]

Totally, this is due to an integration with huggingface to load any datasets from the hub. It requires python and some libraries, everything is supposed to be automatic, but nothing is documented in the case something goes wrong. Also, nothing is in place to create a virtual environment just for burn.

[hscspring]

@mwbryant yes, python3 is needed, not only for the downloader but also the deps installing. Specified, line118 and line196 in the burn-dataset/src/source/huggingfacedownloader.rs should modify to your python, I think it's better to use >=python3.7.

Actually, I think it might no need to do that, a command line flag /path/to/dataset/ might be better.

[John0x]

I think burn would benefit from ditching python all together. I'm experiencing issues with it as well. Not having python involved was the primary reason for my interest in this project :(
I have python 3.11.0 but it's failing to build pyarrow

[nathanielsimard]

Python is only used to download datasets. When it's done, Rust is used to load and process everything data related. I agree that it would be a nice to avoid Python for the default examples.

However, having an integration with hugginface datasets is a plus though, it's an easy access to thousands of datasets with a reasonable Rust API easy to maintain.

[antimora]

I had the same problem when using with Mac. I believe the general problem is that instead of pip, we should refer to pip3. Most systems now use pip3 unless someone explicitly specified aliases like this apt install python-is-python3.

Also we should add some system checks in the dataset package.

[antimora]

This is fixed by #185