aggregate: geometric mean?

[jowens]

This might be a vega issue, not vega-lite, but:

Could you support a geometric-mean? I see "avg", which I assume is an arithmetic mean, and you have "median". I know I'll need geomean in the not too distant future.

[jheer]

In Vega (and the underlying aggregator in datalib), this could be realized via sums of log-transformed values followed by post-aggregate exponentiation. Basically, combining Vega formula and aggregate transforms. (Similar strategies apply to the harmonic mean.) Whether or not we should surface these as high-level options in vega-lite is another question, as we might translate those aggregates to multiple Vega transforms.

Related stack overflow article about SQL: http://stackoverflow.com/questions/3912204/why-is-there-no-product-aggregate-function-in-sql This is also relevant to us as we consider future work partitioning the data flow across client and server.

[jowens]

Fascinating, thanks for the perspective!

It would seem from the API point of view that one of the interesting design points for vega-lite would be how to drop in (call out to?) a Vega feature if that was desired. This would be a good example; I kind-of doubt that vega-lite will support formula and aggregate directly (feel free to tell me I'm wrong), but something in vega-lite syntax that directly "called" Vega would be useful. (In the alternative, I would personally plan to convert vega-lite to Vega and then patch the Vega JSON, but that's not exactly portable.)

[domoritz]

You are totally right. We would like to have a way for users to specify parts of their spec in vega if vega-lite is not expressive enough. However, we have not yet found a clean way to do this in the spec and in the code. I am afraid that this is a major feature that requires some rewriting.

[kanitw]

This is more like a datalib issue, so if you strong feel we should support this now, please file it in the datalib repo. :)

[g3o2]

I've provided a workaround (pow) in the issue referenced here above.