rumbledb / iris-hep-benchmark-rumble Goto Github PK

The hep:histogram function results in offset binning if the bin range doesn't start at a multiple of the bin width. I believe this issue affects your other repositories as well.

https://github.com/RumbleDB/hep-iris-benchmark-jsoniq/blob/45a33842e73cd7fd881e49b81ebc4f4691ff9d97/queries/common/hep.jq#L4-L22

$bucket-idx is set to round(($v - $half-width) div $width). The problem is that this assumes that $half-width is an integer multiple of $width away from $lo. This is equivalent to $lo being an multiple of $width. This is always the case if $lo is zero but is not true generically. When this is not the case, the lowest bin gets shifted to start from the multiple of $width that is closest to $lo.

There are several ways to fix this, but my strategy would be to use floor(($v - $lo) div $width) for $bucket-idx. Then $center := $lo + $bucket-idx * $width + $half-width. Although, it is nice to record the bin edges instead of centers (for example, numpy.histogram()), but that's not as critical.