This repository contains the code from our paper "Online Convex Optimization and Integral Quadratic Constraints: A new approach to regret analysis". The preprint is accessible on arXiv.
The code was developed with Python 3.11. All relevant packages can be installed with
pip install -r requirements.txtAll SDPs are solved with the commericial solver MOSEK.
pip install MosekAn academic license can be requested here. Other open-source solvers might work as well (e.g. cvxopt), however, we observed best numerical stability with MOSEK.
All experiments can be replicated in an associated notebook oco_iqc.ipynb.
🧑💻 Fabian Jakob