Add array aggregation for raw measure metrics, improve testing #282
Conversation
…n/Quantile to return errors for array/ddsketch/maxsumcount
jmacd
requested review from
bogdandrutu,
pjanotti,
rghetia,
sjkaris and
tedsuo
as code owners
| } | ||
| panic("unknown number kind") | ||
| func (n *Numbers) Median() core.Number { | ||
| return n.numbers[len(n.numbers)/2] |
There was a problem hiding this comment.
why is media formula changed?
There was a problem hiding this comment.
I'm using Median() here to mean Quantile(0.5), and there are too many definitions for Quantile to list. (I found http://mathworld.wolfram.com/Quantile.html informative)
After some research in the metrics space, I concluded that the most frequent definition for a quantile returns the smallest value from the original data set which is >= the given quantile. This means we always return a value in the set, not an interpolated value.
Interpolation has risks -- it makes sense when you have a data model, but we don't in this case. If we know the data is uniform or gaussian, then interpolation could make sense.
There was a problem hiding this comment.
I added "// Median() is an alias for Quantile(0.5)." and
// Note that len(n.numbers) is 1 greater than the max element
// index, so dividing by two rounds up. This gives the
// intended definition for Quantile() in tests, which is to
// return the smallest element that is at or above the
// specified quantile.
* Fix non-portable sed invocation in pre_release.sh * Fixes due to API changes * Update CHANGELOG.md
This adds another aggregation type for exporters that wish to see a raw list of values, as opposed to the
maxsumcountandddsketchaggregators, which cannot provide the raw values.Improves the testing framework, adds more DDSketch testing, and adds error values for the Quantile/Min/Max interfaces.