The PDF file contains the extra simulation plots and proofs referenced in the main paper.

The Python file LpBallsCode is used to generate the simulations for lp balls in the paper and Appendix. The parameters r,p,d,N,eta in the code mean the same as they do in the paper. The parameter S is the number of instances to run. In the paper we use either S = 25 or S=10. The parameters can be varied to produce the different plots in the paper. 

There is also an outer p-loop which can be replaced with a d- or r-loop to produce different plots.

Likewise the Python file SchattenBallsCode is used to generate the simulations for Schatten matrix norm balls in the paper and Appendix. The Python file GreedyVSLazy is used to generate Figure 2(b) that suggests Greedy Subgradient is not universal.

We used Python 3.7.9 and the packages listed in the code preamble.
