Ternary Vectorisation Framework

[andrjohns]

Summary

This PR adds a framework for vectorising ternary functions, analogous to that of apply_scalar_binary. It is compatible with all scalar/container combinations, as well as arbitrarily nested containers.

The inc_beta function has been vectorised as an example function

Tests

Testing frameworks for both prim and mix have been added to ensure that both values and gradients are correct when using the vectorisation framework

Side Effects

N/A

Release notes

Added framework for vectorising ternary functions

Checklist

• [x] Math issue #2638

• [x] Copyright holder: Andrew Johnson

  The copyright holder is typically you or your assignee, such as a university or company. By submitting this pull request, the copyright holder is agreeing to the license the submitted work under the following licenses: - Code: BSD 3-clause (https://opensource.org/licenses/BSD-3-Clause) - Documentation: CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/)

• [x] the basic tests are passing

  • unit tests pass (to run, use: ./runTests.py test/unit)
  • header checks pass, (make test-headers)
  • dependencies checks pass, (make test-math-dependencies)
  • docs build, (make doxygen)
  • code passes the built in C++ standards checks (make cpplint)

• [x] the code is written in idiomatic C++ and changes are documented in the doxygen

• [x] the new changes are tested

[andrjohns]

@t4c1 There's an odd failure of the expression tests here:

[ RUN      ] ExpressionTestPrim.poisson_lcdf585
./test/expressions/expression_test_helpers.hpp:101: Failure
Expected equality of these values:
  a
    Which is: -726.65
  b
    Which is: -726.65

Any idea what this could be about?

[andrjohns]

Ah sorry for the ping, figured it out

[bob-carpenter]

figured it out

Now I'm curious. I'm guessing the test uses more precision than the print.

If that's the case, maybe we could change the way it prints by default so this isn't so confusing.

[andrjohns]

Now I'm curious. I'm guessing the test uses more precision than the print.

Yep, there was a numerical difference because one of the captured scalars was falling out scope, but this wasn't shown by the print

[stan-buildbot]

---

Name	Old Result	New Result	Ratio	Performance change( 1 - new / old )
gp_pois_regr/gp_pois_regr.stan	3.56	3.56	1.0	0.15% faster
low_dim_corr_gauss/low_dim_corr_gauss.stan	0.02	0.02	0.98	-1.91% slower
eight_schools/eight_schools.stan	0.09	0.08	1.02	2.01% faster
gp_regr/gp_regr.stan	0.14	0.14	0.99	-0.89% slower
irt_2pl/irt_2pl.stan	5.77	5.69	1.01	1.39% faster
performance.compilation	92.92	91.25	1.02	1.8% faster
low_dim_gauss_mix_collapse/low_dim_gauss_mix_collapse.stan	8.03	8.11	0.99	-0.98% slower
pkpd/one_comp_mm_elim_abs.stan	31.75	32.03	0.99	-0.89% slower
sir/sir.stan	118.17	121.3	0.97	-2.65% slower
gp_regr/gen_gp_data.stan	0.03	0.03	1.0	-0.1% slower
low_dim_gauss_mix/low_dim_gauss_mix.stan	3.02	2.99	1.01	0.7% faster
pkpd/sim_one_comp_mm_elim_abs.stan	0.43	0.41	1.04	4.25% faster
arK/arK.stan	2.06	2.06	1.0	-0.24% slower
arma/arma.stan	0.23	0.23	1.0	0.49% faster
garch/garch.stan	0.58	0.57	1.0	0.48% faster
Mean result: 1.00268204141

---

Jenkins Console Log Blue Ocean Commit hash: e9fa1d783fd4fc7cfe8629346b0d59a726ecae46

---

Machine information

ProductName: Mac OS X ProductVersion: 10.11.6 BuildVersion: 15G22010

CPU: Intel(R) Xeon(R) CPU E5-1680 v2 @ 3.00GHz

G++: Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/usr/include/c++/4.2.1 Apple LLVM version 7.0.2 (clang-700.1.81) Target: x86_64-apple-darwin15.6.0 Thread model: posix

Clang: Apple LLVM version 7.0.2 (clang-700.1.81) Target: x86_64-apple-darwin15.6.0 Thread model: posix

[andrjohns]

Closing for now to clear PR backlog, will re-open once ready for review

[andrjohns]

@serban-nicusor-toptal There's an odd Git failure on the Jenkins run, any chance you've seen something like this before>

[andrjohns]

@SteveBronder this is ready for another look. I ended up adding make_holder to all of the apply_scalar_binary calls, since I was still having issues with scalars falling out of scope

[andrjohns]

@SteveBronder sorry to double-ping, but I'm hoping to introduce an arbitrary vectorisation framework once this is in. Any chance you'd have time for another look?

[syclik]

@andrjohns, thanks! I'm glad we could pull out the additional Eigen plug-in! That simplifies the logic and the burden on developers is significantly lower. I think it'll be worth it in the long run.

There's some additional tech debt added from this PR. It was merged with British spelling of "specialization." I'll put up a PR to address it.

(Yes, choice of American vs British English as the default is arbitrary and wasn't really up for debate, but at this point it's better to be consistent rather than mixed.)