https://arxiv.org/abs/1912.01718
This code will generate all data used in simulations presented in the paper and produce plots. Parameters can be adjusted to produce results for other scenarios.
-
Python v3.xx
-
Modules:
- numpy
- scipy
- matplotlib
- multiprocessing
Executing gen_data.py and light_tailed.py will generate all samples from all distributions given in the paper, and compute estimated CVaRs (alpha=0.998) using CVaR-SA and CVaR-EVT. All samples and CVaR values will be stored in the data folder. Bandit experiment data must be generated in a similar fashion by running cvar_sr_bandit.py and light_tailed_bandit.py. The various plotting scripts will output results stored in the plots folder. Since generating all samples and estimated quantities is quite memory-intensive (requiring the multiprocessing module), all calculated values are already provided in the data folder, but samples must be re-generated by the user due to size constraints.