alg:ChooseX consider changing handling of 0 weight entries

[Implausiblyfun]

Oak should consider either creating a new set of functions for either choosing or stwheap itself to deal with 0 weights.

Problem: When attempting to perform UniqueChooseX from a weighted list and any entry is 0 weight it should never be selected. Today when converting from base weights to cumulative weights we end up with the state where the 0 value entry will be selected whenever the entry after it would be selected. This is annoying when tuning games if you do not want to remove 0 entries.

Decision point 1: Is this a bug and therefore ok to change without a major version change? Or do we need a new set or subset of methods.

Current thought is to have stwheap ignore/immediately pop 0 valuers and or in cumulative values function.

This may also be good time to revist the alg choosing and get better verifications and testing.

[200sc]

I think we can call it a bug if an element with input weight 0 is ever chosen-- we can write some simple tests and add some kind of filtering logic.

[200sc]

I can't replicate this based on the description; @Implausiblyfun Mind writing a failing test?

I wrote this, which passed

func TestUniqueChooseXZeroWeight(t *testing.T) {
    t.Parallel()
    rand.Seed(int64(time.Now().UTC().Nanosecond()))

    for i := 0; i < testCt; i++ { // testCt is one million
        weights, goodIndex := randomZeroWeightSlice()
        chosen := UniqueChooseX(weights, 1)
        if chosen[0] != goodIndex {
            t.Fatalf("choose x chose an element with 0 weight")
        }
    }
}

func randomZeroWeightSlice() ([]float64, int) {
    ct := rand.Intn(20) + 1
    values := make([]float64, ct)
    realVal := rand.Float64()
    realValIndex := rand.Intn(ct)
    values[realValIndex] = realVal
    return values, realValIndex
}

[200sc]

Looking again, I think the bug was the calling code using UniqueChooseX with CumulativeWeights when that function is designed for WeightedChooseOne.

[Implausiblyfun]

Sorry about that should have followed up earlier. I had encountered this on Dashking and I think I mis-described the issue. See the linked PR for what I think I was encountering in that instance. I do want to take a bit more time to check this out and verify its the only oddity but I do think that the main issue that I was encountering where I was getting unexpected entries with 0 weights being selected.