Repository containing the accompanying code to our optimization for machine learning paper "Model Generalization: A Sharpness Aware Optimization Perspective"
Clone the repository
git clone --recurse-submodules https://github.com/CharlieLeee/OptML-Project.git
Set up the anaconda environment with:
conda env create -n OptMLProject --file environment.yml
In order to replicate the experiments in the paper, the folders local_scripts
and izar_scripts
contain the shell scripts that were used to execute the experiments. These expeiments resulted in .csv files containing the results, which are stored in the Notebooks/acc
and Notebooks/loss
folders.
In order to recreate the plots used in the paper, the file Notebooks/Generate plot in the report.ipynb
is used.
izar_scripts
contains the scripts for experiment 1 and 2.
local_scripts/
contain scripts to run the ASAM vs SAM comparison in experiment C.
run.py
contains the code to run the experiments.
Notebooks
contains the experiment data, and the notebook to generate the plots used in the paper.