To find the best expert-based heuristic policy, one must conduct a search over the possible inspection intervals and number of components to be inspected.
This is done via the run_heuristics script.
This script takes as input via its first lines the parameters of the environments and the parameters of the heuristic search.
Execute the script download_heuristic_logs.sh to retrieve the logs of the experiments conducted in the paper.
Reproduce the results: you can either run again the policy search to identify the optimised heuristics or directly evaluate the stored policies.
The policy search can be executed by indicating search = True in the script run_heuristics.py.
To re-run the policy evaluation corresponding to the optimised heuristics, you can directly test the stored policies in heur_search folder. In this case, specify search = False in the script run_heuristics.py.
For example, to compute the return resulting from the uncorrelated 4-out-of-5 environment:
- Check the optimized heuristics: 'insp_interv': 10, 'insp_comp': 5
- The seed was set up as 0 by default
- Execute run_heuristics.py