mk.math.argMax doesn't discriminate doubles between 0 and 1

[occultus73]

I believe I have found a bug with the argMax method. For a Double D1Array like this:
[0.008830892605042162, 0.7638366187431083, 0.2120482416220285, -0.1376875285688708, 0.04677604134217575, -0.11842322358629093, -0.040132636896546836, -0.26975741918728957, 0.006022397547541469, 0.2745433876411484, -0.1882133697860493, 0.0715491465942086, -0.0728390325349591, -0.04587817955186226, -0.3298069957413544, 0.3580707507112211, -0.07260259350541257, -0.0784403405630531, -0.021586318003611993, 0.0557500172047746, 0.02325952046399335, -0.005522260502330234, -0.06656836136167943, 0.16638479724410082, -0.05508238956796159, 0.014393084997982114, 0.03480731825314662, 0.11760367945163475, -0.037000109020572255, -0.053639491632357504, -0.18112447733588335, -0.0024963482748966426, -0.13649951751668346]
It will return zero every time. Only when one of the numbers is above 1, does it return the index of that number, instead of zero.

Presumably the problem is these doubles are being truncated and compared as integers under the hood.

[occultus73]

The culprit appears to be this method:

//TODO(create module utils)
/*internal*/ public operator fun <T : Number> Number.compareTo(other: T): Int = (this - other).toInt()

In org.jetbrains.kotlinx.multik.ndarray.data. It overrides the less-than comparison operator used in the argMax algorithim.

[occultus73]

Gosh that's a surprisingly tricky problem, isn't it? Ideally you could just put in:
(this - other).sign.toInt()

but it's not provided for shorts and bytes:
https://kotlinlang.org/api/latest/jvm/stdlib/kotlin.math/sign.html

[devcrocod]

Thanks for finding this bug.

Kotlin reuses java Double.compare(double d1, double d2), so I think in this case we can use a cast and call compareTo on a specific type. Since there is still a case with NaN.