search

JuliaGaussianProcesses / ApproximateGPs.jl

Approximations for Gaussian processes: sparse variational inducing point approximations, Laplace approximation, ...
https://juliagaussianprocesses.github.io/ApproximateGPs.jl/dev
Other
35 stars 6 forks source link

create ApproximateGPs.TestUtils #117

Closed st-- closed 2 years ago

st-- commented 2 years ago

To make it easier to test various approximations (similar to AbstractGPs.TestUtils)

st-- commented 2 years ago

One thing I notice is that the approx_lml function isn't tested, which you point to in the other PR. Should this at least be checked?

Yeah, we should have that too... maybe later? for now I just wanted to take out the non-EP changes from #64.

Re. my comment about separating out the API from the ability to provide good approximate inference, should we also consider including applying Zygote to approx_lml as part of this test suite somewhere? (I'm not committed to requiring this though).

Testing gradients should probably be there, too ,but also separately (e.g. current EP doesn't have approx_lml implemented, and gettings its gradients right would be another headache on top)

willtebbutt commented 2 years ago

Happy to leave these suggestions for another PR. Please ping me again when you want me to take another look -- I think it's pretty close to being good to go.

st-- commented 2 years ago

@willtebbutt ping

codecov[bot] commented 2 years ago

Codecov Report

Merging #117 (553550f) into master (6a5877f) will decrease coverage by 0.00%. The diff coverage is 94.11%.

@@            Coverage Diff             @@
##           master     #117      +/-   ##
==========================================
- Coverage   94.13%   94.13%   -0.01%     
==========================================
  Files           4        5       +1     
  Lines         290      324      +34     
==========================================
+ Hits          273      305      +32     
- Misses         17       19       +2
Impacted Files Coverage Δ
src/TestUtils.jl 94.11% <94.11%> (ø)

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 6a5877f...553550f. Read the comment docs.