search

SystemsGenetics / KINC

Knowledge Independent Network Construction
MIT License
11 stars 4 forks source link

Fix for long corrpower analytic #132

Closed spficklin closed 4 years ago

spficklin commented 4 years ago

This is a fix for the corrpower analytic. As described in issue #126, this analytic would take an extreme amount of time to complete. The problem was that it was not using a correct index for each working block and would repeatedly process the same elements. To fix this, I added a seek function to the pairwise_matrix_pair.cpp file that is used to properly find the index for each block.

bentsherman commented 4 years ago

@spficklin @4ctrl-alt-del I am concerned that this solution is a bit more complicated than it needs to be. Now that we have a couple of analytics that use MPI (conditional-test, corrpower, similarity), I'm going to review them to see if there is a best practice that will apply to all of them. Then I will get back to you about this PR.

spficklin commented 4 years ago

@bentsherman the cond-test doesn't break under MPI but it's not working properly. It has a similar problem that needs the fix in this PR. See issue #133. The similarity analytic is fine because it works through every cell of the matrix. The cond-test and the corrpower have to deal with a sparse matrix.

spficklin commented 4 years ago

Oh, I should note, The commit that @4ctrl-alt-del just made is to correct issue #134. Thanks @4ctrl-alt-del !

bentsherman commented 4 years ago

I just pushed some changes that should fix the issues you've been having with corrpower and cond-test, so this PR is no longer needed.