protocol_subtomo_subtraction: improved implementation with MPI and md convert #164
Conversation
| @@ -77,87 +75,29 @@ def _defineParams(self, form): | |||
| expertLevel=LEVEL_ADVANCED, help='Save input volume 1 (first subtomogram of the set) filtered ' | |||
| 'and input volume 2 (average) adjusted, which are the volumes ' | |||
| 'that are really subtracted.') | |||
| form.addParallelSection(threads=0, mpi=4) | |||
There was a problem hiding this comment.
Why 4 and not 8? (is the default in other cases)
There was a problem hiding this comment.
Why we can not fill the MPI number in the form?
There was a problem hiding this comment.
Despite I am still learning the algorithm the MPI can overload the RAM. That means at least 8 copies in memory, but the algorithm can requires many more (internally). I think 4 is a save value. But, I am still learning the method
There was a problem hiding this comment.
Yes, and AFAIK 4 was the maximum allowed in the machine where Scipion tests were executed (maybe this has changed now, not sure). But I have no problem if you change it as you consider.
With respect to why we can not fill the MPI number in the form, I could when I tried (actually I tried with different values), so not sure what has changed (maybe something from scipion is not updated or has changed?). I am so sorry but I can not check it now as my Ubuntu has died (I do not know why... the destiny?) but let me know if you need something @albertmena .
There was a problem hiding this comment.
Ok, now it works (not sure what happened)
|
Kudos, SonarCloud Quality Gate passed!
|
This PR goes with I2PC/xmipp#825
The implementation has changed: now it converts the set of subtomograms to a xmd, subtract all of them (doing all the preprocessing needed in xmipp) and then it returns the output xmd which is converted to a setOfSubtomograms.
This development has implied to create a xmipptomo convert (based on xmipp3 convert) with some adaptation to subtomograms and coordinates3D. Probably this convert can be improved in the future.