Default number of bootstraps

[brshallo]

In the paper Bootstrap Prediction Intervals in Non-Parametric Regression... (Dan Saattrup Nielsen's post on implementation, my post mostly mirroring Dan's) which I believe follows essentially the same procedure as workboots, the authors recommend the number of bootstrap simulations to be the square root of the number of samples. Therefore, you might consider changing the default argument in predict_boots() from n = 2000 to n = sqrt(nrow(training_data)).

(Note that Max Kuhn suggests here you may want a large number of simulations, though not sure exactly what the number would be and am guessing it'd be model specific.)