Machine Learning on ARM

I figured Id post on here because I am unclear about a couple things and I figured it would be a quick way to clear up some questions I have.

Im curious if libraries like SK Learn, numpy, pandas, h5py etc. are compatible with ARM based processors?

I see for example that Anaconda appears to have a Linux version available for ARM but I am not clear for example if they differentiate between x64 and ARM64 packages and if there's a difference in whats available etc.

I am also curious about running a Docker image with Anaconda and Jupyter notebook and remotely connecting to it. I would run it on an Oracle Cloud instance (ARM based).

Any feedback from anyone who has some insight on these issues?

Thanks!

  • Pretty much all the tools you mention are just python libraries, so it's just a matter of having a native python build for your OS / aarch64. Linux has that, so if that's your OS, you should get good performance.

    Anaconda also has a Linux/aarch64 package (make sure you find that one, rather than the x64 one)

    I haven't personally done cloud-based, but there shouldn't be any particular issue if you have access to Arm64-based cloud services. Docker has appropriate images.

  • Yes, the mentioned python libraries could run on ARM Cortex A processors with Linux.

    Instead of cloud, you could also run jupyter notebook on the ARM processor and access it remotely via ssh from your PC.
    for eg) this is how I accessed it via my PC while the Juptyer notebook was running on Jetson Nano board.
    github.com/.../jetson_nano_setup_instructions.md