ajwerner

The problem is with collinear points. Adding a small amount of jitter to the point coordinates should prevent collinearity. That being said, it is a bug, if I find some...

Do be aware that a lot of Voronoi libraries carry this same problem as it's the main degenerate case. https://github.com/d3/d3-voronoi/issues/12 was closed with the advice to add jitter.

Yes, the horizontal points should be handled better. In general however floating-point imprecision is a real problem that the original framing of Fortune's algorithm doesn't take in to account. Dealing...

I'm thinking I'd add a new `types.ImportMode`, maybe `ImEmbeddedThirdParty`, ends up being a mashup between `ImThirdParty` and `ImInception` with some differences. It is an external package to both the `imports`...

Yes, that makes sense. I'll give it a shot when I get a shot when I get a chance.

https://github.com/ajwerner/tdigest is goroutine safe and has merging implemented if you want to check it out.

> If events don't have event times or you don't want to respect them then you have to depend on processing time. This is bad juju because it means that...

For some more context, I'd love to hack soon on adding first-class support for t-digests into prometheus (https://github.com/prometheus/prometheus/issues/2682). Prometheus today allows you to report histogram buckets where each bucket is...

I don' have any answers but maybe these thoughts can whet your appetite on the problem (or discourage, who knows). The first step to first class support in prometheus will...

>One possibility is to not use an additive measure if Prometheus allows that. does it? There's nothing at all that bounds us to using an additive measure. That's an assumption...