Calling prior in sampling with fixed parameters does not work

[arrjon]

The NegLogParameterPriors objective does not properly account for fixed parameters when invoked during sampling. When it's integrated into the AggregatedObjective, optimization functions correctly; however, issues arise during sampling because call_unprocessed is triggered with the reduced parameter vector, causing a failure.

Maybe there is a better way to fix this, as I just updated the call of neglogprior, but I feel this might break stuff during optimzation.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 84.52%. Comparing base (049493c) to head (e555ef2).

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1378      +/-   ##
===========================================
+ Coverage    84.48%   84.52%   +0.03%     
===========================================
  Files          157      157              
  Lines        12958    12956       -2     
===========================================
+ Hits         10948    10951       +3     
+ Misses        2010     2005       -5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[arrjon]

I reverted the inital fix since I found a better solution. If you now call problem.x_priors it behaves exactly as problem.objective, which is the behaviour I would expect (wrt fixed parameters etc.).

[dilpath]

Fine for me for now 👍

I imagine this will be obsolete if #1325 is resolved by implementing objective type flags.

[vwiela]

Thanks for spotting.

Think this is a good fix to it.

[arrjon]

After discussion, @PaulJonasJost and I removed the deepcopy of the NegLogParameterPriors class. Otherwise, copying the prior does not work. This happens in the parallel tempering sampler, when the prior is copied to the different chains.

[PaulJonasJost]

Thanks, LGTM