This repo contains the full source code and results for the flatness-guided HPO paper. While this was in development, the method was also called STCVX or AHSC temporarily, so you will likely find files with those names instead.
image/- Experiments on MNIST and SVHNhpobench/- Experiments on the OpenML datasetsbbo_challenge/- Experiments on the Bayesmark datasets
In a newer commit, we have moved the results to an S3 bucket to save space. These are publicly available on s3://stcvx-results/. Because earlier commits had some results, we recommend that you clone using git clone --depth 1 to avoid downloading the results.