Xy block

[dmbates]

• Replace the last two blocks, X and y, in the terms field with a single femat matrix that includes both X and y
• This cuts down on the number of calls to multiplication, etc. in updateL! and allows multi-threaded BLAS to be more effective
• This PR is incomplete

[dmbates]

• Changes in the GLMM code to accommodate the new structures.
• Several changes are because the GLM package specializes some methods to Vector{T}. If those were changed to AbstractVector{T} the changes would not be necessary.
• Some of the changes of arguments like foo.bar.baz' to adjoint(foo.bar.baz) were simply to get the linter to shut up. They were being flagged as syntax errors despite the fact that they were not syntax errors.

[palday]

Thanks for the explanatory comments. :) Should we open a PR against GLM for changing loosening the type requirements?

[dmbates]

@palday I am done tinkering. Now if you could just do the heavy lifting of resolving conflicts, etc.

[codecov]

Codecov Report

Merging #464 (ba2961f) into master (f81ccbc) will decrease coverage by 0.06%.
The diff coverage is 95.32%.

@@            Coverage Diff             @@
##           master     #464      +/-   ##
==========================================
- Coverage   93.49%   93.43%   -0.07%     
==========================================
  Files          27       27              
  Lines        1906     1949      +43     
==========================================
+ Hits         1782     1821      +39     
- Misses        124      128       +4     

Impacted Files	Coverage Δ
src/MixedModels.jl	100.00% <ø> (ø)
src/arraytypes.jl	93.47% <0.00%> (+2.17%)	⬆️
src/blockdescription.jl	100.00% <ø> (ø)
src/generalizedlinearmixedmodel.jl	88.16% <84.61%> (-0.07%)	⬇️
src/Xymat.jl	86.84% <86.84%> (ø)
src/linalg.jl	59.72% <100.00%> (ø)
src/linalg/logdet.jl	100.00% <100.00%> (ø)
src/linearmixedmodel.jl	95.01% <100.00%> (+0.70%)	⬆️
src/mixedmodel.jl	83.33% <100.00%> (ø)
src/remat.jl	95.35% <100.00%> (-0.02%)	⬇️
... and 5 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f81ccbc...63c32db. Read the comment docs.

[palday]

The failure on Future is something I've fixed as part of my changes in #474

[dmbates]

In v"1.7.0-DEV.620" and later, which use libblastrampoline, the first use of BLAS.vendor() produces a deprecation warning. I can commit a change for that if it seems worthwhile.

[palday]

Go for it -- we've already got a @static guard there noting that I would still like to have tests notify us of platform specifics (since those have been the source of problems in the past).

[dmbates]

I think we should look at pivotedQR versus pivotedCholesky for determining the rank of X after we merge this. So probably don't make a new release immediately.

[palday]

I also think there's a few things to think about before we put a bow on 4.0.

[palday]

There's also something I'm currently tracking down with the new show methods. In my local build of the docs, the markdown is used and renders correctly (not what we want for showing users how to step through the REPL, but also at least understandable). But the GHA builds all have the markdown output but not rendered into HTML. I'm working on figuring that out.

[dmbates]

@palday Let me know if I should be looking at this PR or if you are ready to do the heavy git lifting to make it compatible with current master.

[palday]

I already did the lifting. 😄

[dmbates]

I think this is ready to go. If you agree @palday please go ahead and merge.