search

philippabele / deep-groove-with-pytorch

0 stars 1 forks source link

Try out AutoML for classification #33

Closed philippabele closed 1 year ago

philippabele commented 1 year ago

After developing a model and performing a parameter study on your own, it's time for using some frameworks to support your machine learning journey.

Take a look at Auto-PyTorch

And try a basic AutoML Search for finding a better network architecture and well selected hyperparameters.

There is an example for tabular data with classification in the projects repository.

https://github.com/automl/Auto-PyTorch/blob/56a2ac1d69c7c61a847c678879a67f5d3672b3e8/examples/20_basics/example_tabular_classification.py

Focus on getting the AutoML process running. I will add an issue for documentation.

RicoKursidem commented 1 year ago

Hi @philippabele, The autoPyTorch Package does not support Windows. I only work on Windows and don't have a Linux Computer. Should I try to run it on a Linux VM or set up a Computer?

Source: https://github.com/automl/Auto-PyTorch/issues/385#issuecomment-1128569008

philippabele commented 1 year ago

@RicoKursidem Do you know Docker?

You can run a plain Ubuntu Docker image with

Docker run Ubuntu bash

And install all needed tools inside the container. If you like, you could also connect vs code to work with it.

If you like. I can give you a live demo via web meeting. Just let me know.

If you are familiar with other solutions like VM feel free to select your favorite way.

RicoKursidem commented 1 year ago

@philippabele

I'm familiar with docker, but I only used it to host Applications on my server. I would like to try it via docker and connect it with VS code. Is it possible to make a meeting tomorrow or Friday?

philippabele commented 1 year ago

@RicoKursidem

Please check your mail inbox 📥

RicoKursidem commented 1 year ago

Hi @philippabele,

To run the docker container and the autoML application is not as straight forward as I thought. I can load the python environment and install the packages given in the autoML Installation guide. But there are Dependencies I cannot find and other errors. I guess the Dependencies are there on a normal Ubuntu installation, but not on the docker container, but I did not test this on a full Linux system.

I tried to run the script from the tutorial, so it's not my code braking the program, but it did not work.

Did you set up a python Environment before? Do you have tips for me? Does the Ubuntu Image contain a full Ubuntu Image or a lighter version?

philippabele commented 1 year ago

@RicoKursidem Did you notice the Dockerfile on the Auto-Pytorch Repository?

https://github.com/automl/Auto-PyTorch/blob/master/Dockerfile

I looks like they faced similar issues.