Description
The behaviour of mine_clique_embeddings seems to be independent of the argument mask_bound, whereas we would expect the routine to become slower, and produce a higher quality cache (on average) when using large values.
To Reproduce
qpu = DWaveSampler()
G = qpu.to_networkx_graph()
BCquick = mine_clique_embeddings(G, mask_bound=1, heuristic_bound=0, num_seeds = 1)
BCslow = mine_clique_embeddings(G, mask_bound=2**8, heuristic_bound=0, num_seeds = 1)
In this second, case we anticipate a much slower search (learding order linear in mask_bound), but instead it is independent of this argument.
Description The behaviour of mine_clique_embeddings seems to be independent of the argument mask_bound, whereas we would expect the routine to become slower, and produce a higher quality cache (on average) when using large values.
To Reproduce qpu = DWaveSampler() G = qpu.to_networkx_graph()
BCquick = mine_clique_embeddings(G, mask_bound=1, heuristic_bound=0, num_seeds = 1) BCslow = mine_clique_embeddings(G, mask_bound=2**8, heuristic_bound=0, num_seeds = 1) In this second, case we anticipate a much slower search (learding order linear in mask_bound), but instead it is independent of this argument.
Workaround (suggested by @boothby)
minorminer.busclique.busgraph_cache.clear_all_caches() BCslow = mine_clique_embeddings(G, mask_bound=2**8, heuristic_bound=0, num_seeds = 1)
Expected behavior Runtime scales with mask_bound, quality of cache increases on average
Environment:
Additional context Add any other context about the problem here.