./display.py ../data/scen_30/1.yaml -a -nbest -f5
where a means animation, best is the name of the solution and f is for making the animation run 5 times real time.
./search.py -m ex ../data/scen_small/8_1.yaml
performs an exhaustive search (tries all solutions). -m ex2 does the same in C++, it's faster
./search.py -m sa ../data/scen_30/1.yaml
performs a simulated annealing run. sa2 and sa3 are C++ equivalent
Same as above but uses python multiprocessing to run searches in parallel
./psearch.py ../data/scen_30/1.yaml -e 50k -t 12 -J 8
This runs 12 simulated annealing searches, each 50k episodes. Searches will be run in parallel batches of 8. If unspecified, the number of available CPU cores will be used. On my laptop, it runs in 1s and almost always returns what I believe to be the optimal solution (82.558s, [08-15-28-07-11...]).
A tool for running and saving multiple searches.
./tune_hp.py -c -s ../data/scen_small/15_6.yaml -t 10 -e 1k,5k /tmp/scen_15_6_runs_10.npz
runs and store the search results, ie 10 runs for 1k episodes and 10 runs for 5k episodes.
./tune_hp.py -x /tmp/scen_15_6_runs_10.npz
load previously computed results and displays histograms
install cython, move to src/native directory, run make, add the native directory to your LD_LIBRARY_PATH (so that libpm.so can be found)
Search methods sa2, sa3 and he2 require building the native code, he and sa don't.