search

BolognaBiocomp / deepsig

DeepSig - Predictor of signal peptides in proteins based on deep learning
https://deepsig.biocomp.unibo.it
GNU General Public License v3.0
25 stars 4 forks source link

More CPUs used than specified via --threads #28

Closed oschwengers closed 2 years ago

oschwengers commented 2 years ago

Hi @savojard , I benchmarked version 1.2.3 to find the sweatspot in terms of provided threads. Unfortunately, deepsig uses all available CPUs on a given machine during the signal peptides detection step. The cleavage site detection seems to work as expected.

For example executing deepsing with --threads 8 still used 8 CPUs which was the max on this machine. This will result in very problematic situations if multiple processes of deepsig run in parallel, e.g. on an HPC systems. Could you please try to fix this behavior?

oschwengers commented 2 years ago

Sry, I just recognized that the above numbers are of course wrong and misleading.

On a machine with 8 cores I executed DeepSig with --treads 4. Unfortunately, DeepSig still allocates 8 threads and uses all 8 cores for the signal peptide detection step. biocrf-static works as expected with a max of 4 threads in this case. Any ideas how to fix that?

savojard commented 2 years ago

Hi @oschwengers

Setting Tensorflow parallelism is not a trivial task. However, I attempted a fix and drafted a new release 1.2.4. Please, let me know if its working. Thank you!

oschwengers commented 2 years ago

Hi @savojard , yes, it works like a charm now (installed via Conda). Thanks a lot!