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
If I run this interactively in a Jupyter Notebook on 1 core I get 80% of the way through the process in 4 hours (the maximum amount of time I'm able to run an interactive session for) at 18.05s/iteration. If I submit the same data/script but as a HPC job utilising 6 cores the job runs significantly more slowly at 464.09s/iteration progressing about 5% of the way through in 1 hour. Both are running the same conda environment on the same object on the same hardware
This is my environment, running on CentOS Linux 8.1
are you sure the jupyter notebook is only using one core (i.e. did you verify in e.g. htop)? Not all schedulers enforce that.
have you considered the tcrdist metric instead of alignment? It should give very similar results while being much faster.
lastly, I'd suggest to upgrade to v0.18. It doesn't affect the speed of the alignment metric, but the subsequent clonotype clustering step is much faster now.
Describe the bug
I'm trying to calculate the IR distance for 100k TCR sequences:
If I run this interactively in a Jupyter Notebook on 1 core I get 80% of the way through the process in 4 hours (the maximum amount of time I'm able to run an interactive session for) at 18.05s/iteration. If I submit the same data/script but as a HPC job utilising 6 cores the job runs significantly more slowly at 464.09s/iteration progressing about 5% of the way through in 1 hour. Both are running the same conda environment on the same object on the same hardware
This is my environment, running on CentOS Linux 8.1
Could you offer any advice?
Thanks!
The text was updated successfully, but these errors were encountered: