Deflation for CCA and LDA

[codecov[bot]]

Codecov Report

Merging #46 into master will decrease coverage by 13.16%. The diff coverage is 55.5%.

@@            Coverage Diff            @@
##           master    #46       +/-   ##
=========================================
- Coverage   90.17%    77%   -13.17%     
=========================================
  Files          32     34        +2     
  Lines        2372   3558     +1186     
=========================================
+ Hits         2139   2740      +601     
- Misses        233    818      +585

Impacted Files	Coverage Δ
R/moma_solve.R	98.07% <ø> (ø)
src/moma.h	100% <ø> (ø)	:arrow_up:
src/moma_solver.h	100% <ø> (ø)	:arrow_up:
src/moma_base.h	100% <ø> (ø)	:arrow_up:
src/moma_solver_BICsearch.cpp	100% <100%> (ø)	:arrow_up:
R/util.R	100% <100%> (+4.54%)	:arrow_up:
src/moma_solver.cpp	94.73% <100%> (ø)	:arrow_up:
R/moma_sflda.R	44.74% <44.74%> (ø)
R/moma_sfpca.R	70.62% <47.59%> (-20.02%)	:arrow_down:
R/moma_sfcca.R	56% <56%> (ø)
... and 8 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 00839b5...ff6da69. Read the comment docs.

[michaelweylandt]

The CodeCov report is excellent! Thanks for adding.

Can we add CodeCov first as a small separate PR so we get reports for all the big PRs before we merge them?

[michaelweylandt]

It looks like some of the commits got misordered: I'm not sure how https://github.com/michaelweylandt/MoMA/pull/46/commits/91b8210eef2c1953bbfbb171e4a9952c8044acdd can follow https://github.com/michaelweylandt/MoMA/pull/46/commits/76280ebc587fc257758d16067dbe3afd7d2cc7fd since the former uses the two argument form of MoMA which is only added in the latter.

[michaelweylandt]

This can now be rebased (not merged) against the current master.

[michaelweylandt]

One thought re:testing (e.g., https://github.com/michaelweylandt/MoMA/pull/46/commits/31ef3c1ff61ec19cf72b278192e541d620b47c6a): in addition to testing that you get the same results from an R and C++ implementation, you can also test that the theoretical properties hold: e.g., if doing Hotelling deflation with regularization, we would expect u^T X_defl v = 0, but not u^T X_defl = 0, while both should be zero for Schur and projection deflation.

It's not the strictest test in the world, but it's at least a different thing to test which is often a good complement to just checking that you coded the same thing in two languages.

[Banana1530]

Good idea. Added.

[michaelweylandt]

Great - just an FYI: @agenevera and I are working on systemitizing the deflation / orthogonality related stuff, so let's focus on the other two open PRs for the next few days. I'm gonna try to write up our thoughts before I go out of town Saturday, but I'm not sure exactly when that will be done by.

[Banana1530]

Gotcha.

[michaelweylandt]

• [ ] This needs to be merged into develop under the new branch scheme instead of master
• [ ] Can we increase coverage of the new stuff - this is a big hit compared to the previous version.

[michaelweylandt]

Can you pull the Shiny stuff out of this PR and just leave it as CCA and LDA? That might help the coverage numbers.

Also, can you clean up the history? It should be possible to move the typo fixes into the original commits for a shorter history.

[Banana1530]

It is moved to PR #54.