The following discussion from !124 (merged) should be addressed:
@rmatev started a discussion: (+6 comments)
I guess convergence is something that we expect from all tests, so instead of creating a new one, shouldn't we just add the validator to the existing tests?
changed title from Implement convergence tests for different solvers to Implement convergence test for different solvers
I was thinking it can be hard to test convergence with less events but maybe enough for this is just to run a few iterations and the validator should check if the last one is done which ensures that several iterations are possible and it doesn't get stuck in any of them. To make it simpler and more useful.
mentioned in merge request !124 (merged)
I am not sure if this was clear, but I added a script: https://gitlab.cern.ch/lhcb/Alignment/-/blob/alignmentmodernizationMay2020/Alignment/Escher/scripts/solvertest.py which runs once over the dataset and then in the end calls different solvers. This will not test convergence for different solvers, but it will at least spit out some information that should allow us to flag if one of the solvers breaks. What I did not do yet, is add a test that actually calls the script.
Yes that's good, it's just when I tried to run the script it wouldn't run in several iterations. It produced the alignlogs for the solvers which can be used to check different things in the test I suppose.
mentioned in merge request !125 (merged)
mentioned in merge request !143 (merged)