violin plots are rendered with long tails in current stable GGPLOT2

[nbafrank]

Dear Hadley,

I and others have notice that the current version of GGPLOT2 renders violin plots with long tails at the upper end unlike previous versions.

Is there any way to revert back to the old rendering?

Thanks,

Francesco

[thomasp85]

Can you provide a reproducible example?

[nbafrank]

res <- data.frame(a = c(1:100),b=c(rep(0,50),rep(1,50)))
ggplot(res,aes(y=a,x=factor(b))) + geom_violin(trim=F)

When you plot the violins, the tails go to reach the very end of the plot (both ends). Prior to version 2.0.0 (I believe) this was not occurring. It is solely an aesthetic change but I was wondering if there's an option to make it look like it used to be.

Thanks,

FV

[nbafrank]

Let me know if you need anything else =)

[hadley]

Use trim = TRUE?

[nbafrank]

so that option will trim and cut the violin. I want a full violin without the ends going all the way to the top like it used to be in the past. Is there a hidden option or you have changed some internal graphic component in an unfixable way? Thanks,FV

[hadley]

Can you please carefully explain the previous behaviour and the current behaviour? I don't understand what you want.

[wch]

@nbafrank in the future screenshots would help. You can use devtools::install_version to install an old version of a package.

This is the plot with 1.0.1:

And the current dev version:

[nbafrank]

Thanks WCH, I did not know about devtools::install_version =) hope this is clear.

[hadley]

@wch do you know what caused the change?

[wch]

Nope, sorry. It looks like the y-range is now determined by the raw data instead of the transformed data.

[hadley]

I guess someone will have to git bisect this

[wch]

@hadley it's your fault: d58b00a

[hadley]

The reason I have an intern is to fix my mistakes...

[thomasp85]

You're going to be in so mush trouble in 7 days

[nbafrank]

@hadley @thomasp85 @wch thanks for your help guys. Looking forward to the new version. Maybe make it an option for both styles?

[thomasp85]

@hadley thinking about this, there is actually a value in the new approach even though it can look a bit odd, in that it communicate the range of the data clearly - do you want a complete roll-back to the old style (without reemerging the bug that caused the change) or what do you think..?

[hadley]

The current dev version doesn't look right because it's truncated to early on one end and too late on another - I think it's probably just that the ranges need to be calculated per group, not overall.