ENH: Add standard errors to Geom.smooth?

[nickeubank]

How hard would it be to add standard errors (the se keyword) to Geom.smooth? Linear and loess fits are something I use a lot, but they're not super useful for inference without confidence intervals.

[Mattriks]

Confidence intervals were added 5 days ago for Stat.smooth(method=:lm) (see #1231). You can use it now by doing add Gadfly#master. For Loess intervals, work needs to be done in Loess.jl first.

[nickeubank]

Oh, great! Thanks @Mattriks

One API question: I'm not sure how closely this packing is hoping to hew to ggplot2, but this doesn't seem to implement the se keyword (I guess it's implied by levels). That deliberate? Works for me either way, just wanted to flag.

[Mattriks]

This is a good example of where Gadfly and ggplot2 differ.
Gadfly uses constructions such as this:

 layer(Stat.smooth(method=:lm, levels=[0.95, 0.99]), Geom.line, Geom.ribbon))

which says that in that layer both Geom.line and Geom.ribbon should use the output from Stat.smooth. Note that Geom.smooth = Geom.line + Stat.smooth, so you can use Geom.smooth if you want the line and no ribbon.

[nickeubank]

oh great, that's very helpful, thanks!