This framework allows continuation for the fidelities data subsets and epochs in the field of multi-fidelity hyperparameter optimization. It allows for various experimentation and thorough result logging as well as their visualization.
It is a project created during the Deep Learning Lab course 2022 of the Albert-Ludwigs-Universität Freiburg.
git clone https://github.com/Bronzila/warmstarting-with-fidelities.git
cd warmstarting-with-fidelities
conda create -n warmstarting python=3.7
conda activate warmstarting
# Install for usage
pip install -r requirements.txt
# Run an experiment defined in config.yml
python run.py --config=config.yml
Define your own experimentation configs or use the predefined in the experiments folder.
Use the run.py file to execute your experiments.
To visualize your results use the functions defined in warmstarting/visualization/plots.py.