Tests for training on subset of data

[pcarbo]

@eweine Can you please add a testthat test (or tests) checking for correctness with training_frac < 1?

[eweine]

Yes!

I believe that the only straightforward way to do this is to fix U and compare fitting V with training_frac = 1 and training_frac < 1. The results should be the same. I will try this.

[eweine]

Hmm I am now realizing that this isn't straightforward with the current way that we have written the code. There is no option to fix U or V, and while you can provide covariates (i.e. X and Z), I don't think you can fit a model with K = 0 and just values of X and Z. @pcarbo should we somehow allow the user to do this? Or, should we figure out another way to test this?

[pcarbo]

If you run with, say, training_frac = 0.99 and training_frac = 1, the results should be very similar on a simple simulated data set (and if you initialize in the same way)?