no-information rate

[topepo]

The computations do a straightforward translation of what is in the 632+ paper. A few months ago I emailed Brad Efron about this asking if it was roughly equivalent to doing a few permutations of the outcome and computing the metric. He responded in the affirmative. I'm going to be implementing the 632+ rule in tune and will be using that approach (because some metrics are not the sum of their single values).

You might think about that too since the current computation is nxn. outer() can help but permutations might be more efficient otherwise.