@mlin, you might be interested to see where this is going.
It seems to be helping seqwish run a little bit faster. However, the model isn't always helping at the scales I'm working at. In the case of the alignment set, iitii improves performance. But this wasn't as much the case for the graph model that seqwish is using. Those might not have the right structure for efficient interpolation. I'll try using the base iit and see if that improves things again.
I see only a handful more optimizations that are worth pursuing. The biggest one is during the transitive closure to build the sequence of the graph, I'm querying for every bp in the entire input sequence set. I should speed things up by using the alignment match information to skip over regions that are homologous.
@mlin, you might be interested to see where this is going.
It seems to be helping seqwish run a little bit faster. However, the model isn't always helping at the scales I'm working at. In the case of the alignment set, iitii improves performance. But this wasn't as much the case for the graph model that seqwish is using. Those might not have the right structure for efficient interpolation. I'll try using the base iit and see if that improves things again.
I see only a handful more optimizations that are worth pursuing. The biggest one is during the transitive closure to build the sequence of the graph, I'm querying for every bp in the entire input sequence set. I should speed things up by using the alignment match information to skip over regions that are homologous.