You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for this package. I have run the development release successfully on a few tumor samples. I have lately been running into persistent memory issues especially in phylogeny steps. I am running on our lab's local server with 64 Gb RAM and 8 cores. I have begun to suspect that the issue is the use of tidyverse in mclapply routines. I have recently converted every mclapply step into a basic for loop assigning to a pre-allocated list--ignoring any parallelization speed benefits. That has helped a little bit. I have considered trying to use dtplyr to ease some memory issues.
Do you think this suspicion is reasonable? Have you already made efforts to reduce the memory footprint?
The text was updated successfully, but these errors were encountered:
I was thinking about trying something like this, but I didn't have the time.
The correct approach above is if you create a pull request with those changes for us to review----please provide statistics on memory usages improvements as well if you could.
After forking the repo, please use the develop branch for these changes, and create a PR against that branch.
In the most recent update 0.1.3, we have completely replaced mclapply in the phylogeny part using RcppParallel. This led to a 10-20x speedup and the memory usage should now stay constant with respect to the number of threads. Please let us know if that helps with your memory issue!
Thanks for this package. I have run the development release successfully on a few tumor samples. I have lately been running into persistent memory issues especially in phylogeny steps. I am running on our lab's local server with 64 Gb RAM and 8 cores. I have begun to suspect that the issue is the use of tidyverse in mclapply routines. I have recently converted every mclapply step into a basic for loop assigning to a pre-allocated list--ignoring any parallelization speed benefits. That has helped a little bit. I have considered trying to use dtplyr to ease some memory issues.
Do you think this suspicion is reasonable? Have you already made efforts to reduce the memory footprint?
The text was updated successfully, but these errors were encountered: