It turned out that in fact the KHash set was optimized over the KHash map, so adding abundancies increases the memory requirements by ~10%. However, once in the parametrization of k-mer size types, I could also try to use sets when m is set to 1.
We could also use an upper bound for k-mer frequencies (as in Jellyfish) as it's quite simple and the slowdown it intriduces should be marginal , but it depends on how useful this in fact is.
It turned out that in fact the KHash set was optimized over the KHash map, so adding abundancies increases the memory requirements by ~10%. However, once in the parametrization of k-mer size types, I could also try to use sets when
m
is set to 1.We could also use an upper bound for k-mer frequencies (as in Jellyfish) as it's quite simple and the slowdown it intriduces should be marginal , but it depends on how useful this in fact is.