Readme

Documentation Status pre-commit.ci status Python package Check Markdown links

DEPRECATED: PLEASE USE https://github.com/gnn-tracking/hyperparameter_optimization2 INSTEAD

This repository hosts submission scripts and framework for hyperparameter optimization of the models defined in the main library. Part of this are fully parameterized versions of the models.

Framework

• Uses ray tune as overarching framework. For deployment on SLURM managed HPC nodes, ray workers are deployed as SLURM batch jobs (as further described here)

• Optuna is used to power the search

• Results are reported to wandb/weights & biases

Setup

First, follow the instructions from the main library to set up the conda environment and install the package

pip install -e .
git submodule update --init --recursive

Get started

• Use or adapt one of the tuning scripts in scripts/

Training with fixed parameters (no tuning)

Other links

• ray-tune-slurm-demo: Simple project to try out some of the capabilities of ray tune and wandb, especially with batch submission

• wandb-osh: package to trigger wandb syncs on compute nodes without internet

• additional stoppers for ray tune: package with additional early stopping conditions for trials used in our HPO