A Comprehensive Framework for Benchmarking Algorithms Across Hyperparameter Spaces

Ammar Gilani, Han Liu

Keywords: Benchmarking, Hyperparameter optimization

Abstract:

We introduce a framework for benchmarking algorithms with varying hyperparameters from multiple perspectives. The dependency of algorithms' performance on hyperparameters complicates fair comparisons and often leads to inconsistent empirical studies. Our framework addresses this challenge by proposing two key criteria: \textit{Performance-HPO} trajectory and \textit{Reliability-HPO}.
The Performance-HPO trajectory tracks how an algorithm’s performance changes with different hyperparameter optimization (HPO) budget allocations, leveraging a variety of off-the-shelf hyperparameter optimizers. This enables users to identify the most suitable algorithm for their specific needs. The Reliability-HPO criterion evaluates the expected value of an algorithm's success rate across hyperparameters, estimated using Monte Carlo simulations in log-space.
We demonstrate our framework by benchmarking widely-used convex optimizers. Our experiments, conducted with {\footnotesize\texttt{CVXPY}} across various problem types, settings, and dimensionalities, reveal that the {\footnotesize\texttt{SCS}} solver exhibits the highest Performance-HPO, while {\footnotesize\texttt{ECOS}} and {\footnotesize\texttt{MOSEK}} demonstrate superior Reliability-HPO.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 11511