Remove GLSFitter dependence from Residuals

[abhisrkckl]

This requires a bit of code duplication, but I think it is worth decoupling these modules. Also, this avoids a bunch of unnecessary fitting steps while computing the chisq.

[codecov]

Codecov Report

Patch coverage: 87.50% and no project coverage change.

Comparison is base (dbe69cc) 68.24% compared to head (72d817f) 68.24%.
Report is 19 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1618   +/-   ##
=======================================
  Coverage   68.24%   68.24%           
=======================================
  Files          99       99           
  Lines       22786    22806   +20     
  Branches     3918     3920    +2     
=======================================
+ Hits        15551    15565   +14     
- Misses       6274     6278    +4     
- Partials      961      963    +2     

Files Changed	Coverage Δ
src/pint/residuals.py	74.81% <78.78%> (+0.27%)	⬆️
src/pint/fitter.py	83.69% <100.00%> (-0.37%)	⬇️
src/pint/models/timing_model.py	83.62% <100.00%> (+0.05%)	⬆️
src/pint/utils.py	60.10% <100.00%> (+0.54%)	⬆️

... and 1 file with indirect coverage changes

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

[dlakaplan]

I assume there is no big performance hit when using this? Or maybe it's even faster?

[abhisrkckl]

It's the same code that was in GLSFitter, sans a few unnecessary lines. There is no reason it should be slower.

[dlakaplan]

I agree that there is no reason. and it probably does less copying of objects. but I just wanted to verify empirically.

[abhisrkckl]

I did the profiling. The old code takes about 4 s whereas the new code takes about 1 s.

[dlakaplan]

great! that's what I was hoping.

[abhisrkckl]

There is an API change in this that has to be documented.

[dlakaplan]

Is this ready to merge?

[dlakaplan]

I guess on the PINT call last week we were discussing whether or not this would lead to changes in any example notebooks. Have you looked to see what needs to be done?

[abhisrkckl]

I checked. The change doesn't affect any notebooks.

I have updated the docstrings to reflect the API change.

[abhisrkckl]

Is this ready to merge?

Yes.

[dlakaplan]

But the understanding_fitters notebook does:

# Not sure how to do this properly yet.
# glsfit2 = pint.fitter.GLSFitter(toas=t, model=glsfit.model, residuals=glsfit.resids)
# glsfit2.fit_toas(maxiter=0)

presumably this is trying to construct residuals and just doing a fit because the API required it?

[abhisrkckl]

I'll check this.

[abhisrkckl]

But the understanding_fitters notebook does:

I don't think those commented out lines are needed. I have removed them.

The API change is only in Residuals and not GLSFitter. The understanding_fitters notebook does not use Residuals directly at all. So I don't think we need any other change in this file.