Storage of (large) test data

[athowes]

Now for some data sets used in tests we are creating them with inst/generate_examples.R then storing them in inst/extdata. At the moment this is fit.rds and fit_gamma.rds. These files are above the 50.00 MB that GitHub recommends as maximum file size.

One option is to thin down these fits to make them smaller.

Another option is that this is the wrong approach and we should rethink where we store data for tests / how we are approaching this somehow.

[seabbs]

One option is to thin down these fits to make them smaller.

In the first instance thin down and yes its definitely a "wrong" approach but I think it works for now. We should have an issue to explore alternatives. IMO I can't see a reason these would need to be this big

[athowes]

I'm happy with closing this issue on thinning then adding another issue for alternatives.

[seabbs]

I just looked and it seems like we could largely replace this approach by running a model fit in setup.R?

[seabbs]

Note that setup.R always runs before the test suite runs and so anything in it is always available.

[athowes]

The downside is that when I am doing things locally I run setup.R to generate the objects needed. So ideally there wouldn't be long running things in there (like model fitting)

[seabbs]

but surely any fit we need for a test is going to be <30 seconds? There really doesn't seem like a need for more?

[athowes]

The Gamma one is not <30 seconds currently.

And If we are intending to do "parameter recovery" integration tests then they can't be bad fits.

[seabbs]

That doesn't seem ideal and it feels like the model should be workable with 30 seconds a core so I find this surprising?

[athowes]

I might not have been setting cores as argument. I can post here what the actual runtimes are.