-
Notifications
You must be signed in to change notification settings - Fork 5
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
TestUtils #18
Comments
Totally agree! I also think the 3 things you mentioned are the main points to cover. We could also start testing different ADs like in kernel functions. I think this needs to be done in AbstractGPs too. |
Yeah, we really need to figure out the "correct" way to test AD in general. Currently everyone writes their own code to test a given AD, which is just bad -- the ADs should provide utility to do this. For now we're probably stuck with code duplication though. edit: by correct, I mean a way that covers all of the technical points correctly, and maximises code reuse between packages. |
Sounds like this is solved no? |
I think before implementing any more likelihoods we should think carefully about what standardised tests we want to run on them, similarly to what I've just done in this PR: JuliaGaussianProcesses/KernelFunctions.jl#159
Looking at the various open PRs (#17, #16 , #15, and maybe #14), and the already implemented likelihoods (
Gaussian
andPoisson
) there appears to be quite a lot of repeated code in the tests. It would be great to get a handle on this before it becomes too large a job to easily refactor.Pulling out some common themes from these, it looks like we want to be testing that
@sharanry you've done most of the work on this so far. Is there anything else that's obvious to you that we're missing?
There's also quite a lot of code that manually generates input and output data / constructs appropriate "latent"
AbstractGPs
at the minute -- in my opinion it would be a very good idea to factor this code out as well.The text was updated successfully, but these errors were encountered: