Incorrect results in R 2.15

[Kasi]

Hi,

  I just updated to 2.15 and reinstalled the ggplot library. Now facet_grid() displays incorrect data. Does anyone else have this problem? Just running the first example listed here (http://had.co.nz/ggplot2/facet_grid.html) already shows a slightly different graph.

I use ggplot2_0.9.0.

Thanks.

[Tyler Rinker]

Without any code to reproduce your problem I'm only guessing at the solution.  It really is very helpful if you display some sample data and code that recreates the error.

I think this may be of use:

http://stackoverflow.com/questions/9778232/ggplot-unexplained-outcome 

Tyler

---

Date: Thu, 5 Apr 2012 15:04:55 -0700
From: pop...@gmx.net
To: ggp...@googlegroups.com
Subject: Incorrect results in R 2.15

--
You received this message because you are subscribed to the ggplot2 mailing list.
Please provide a reproducible example: http://gist.github.com/270442
 
To post: email ggp...@googlegroups.com
To unsubscribe: email ggplot2+u...@googlegroups.com
More options: http://groups.google.com/group/ggplot2

[Kasi]

It's very simple to reproduce. Run this:

library("ggplot2")
p <- ggplot(diamonds, aes(carat, ..density..)) + geom_histogram(binwidth = 0.2)
p + facet_grid(. ~ cut)

And compare it to the first graph on this page:

http://had.co.nz/ggplot2/facet_grid.html

If they are not the same, then either the diamonds data set has changed or somewhere the is a bug.

[Tyler Rinker]

I run exactly what you have and get this:

http://dl.dropbox.com/u/61803503/ggplot.op.png

I don't see a difference between this and the link you supplied.  Unless there's something I'm missing I'd suggest you try rebooting R as perhaps you've overwritten something that's related/necessary for this plot.

---

Date: Thu, 5 Apr 2012 15:55:29 -0700
From: pop...@gmx.net
To: ggp...@googlegroups.com
CC: pop...@gmx.net
Subject: Re: Incorrect results in R 2.15

[Tyler Rinker]

I just happened to notice that there are slight differences in the histograms I posted and the one from the website you posted.  This may be due to a slight change in binwidth or the graphics device used (mine is a screen shot of Windows R gui).   It is hard to know what your problem is as you haven't been explicit in what you believe is incorrect.  Could you explain a bit more about the discrepancy you're seeing?

Tyler

---

From: tyler_...@hotmail.com
To: pop...@gmx.net; ggp...@googlegroups.com
Subject: RE: Incorrect results in R 2.15
Date: Thu, 5 Apr 2012 19:14:27 -0400

[Jean-Olivier Irisson]

On 2012-Apr-06, at 01:28 , Tyler Rinker wrote:
>
> I just happened to notice that there are slight differences in the histograms I posted and the one from the website you posted. This may be due to a slight change in binwidth or the graphics device used (mine is a screen shot of Windows R gui). It is hard to know what your problem is as you haven't been explicit in what you believe is incorrect. Could you explain a bit more about the discrepancy you're seeing?

I also see differences in R 2.14 so I think the change comes from ggplot 0.9 vs 0.8.9 rather than from R 2.14 vs R 2.15.
It's not a graphic device or platform thing either (and it shouldn't be) since I see the same graph Tyler posted but I am on OS X with a quartz device.

I suspected stat_bin changed the way the bounds of the successive bins are included/excluded. And indeed there is this in the commit log:

commit bb4ec84bbedadd95647760fe586f64d0d6216abb
Author: hadley <h.wi...@gmail.com>
Date: Wed Dec 21 08:12:24 2011 -0600

Default to right-open, left-closed intervals in hist (Fixes #194)

So I checked :

library("devtools")
# install old version of plyr, compatible with the old ggplot2
install_github("plyr", branch="plyr-1.6")
# install previous commit
install_github("ggplot2", branch="1bfbee0fc791de8bfef901c156ebd04c4765b194")
library("ggplot2")
ggplot(diamonds, aes(carat, ..density..)) + geom_histogram(binwidth = 0.2) + facet_grid(.~cut)
# install suspected commit
install_github("ggplot2", branch="bb4ec84bbedadd95647760fe586f64d0d6216abb")
ggplot(diamonds, aes(carat, ..density..)) + geom_histogram(binwidth = 0.2) + facet_grid(.~cut)

and indeed, this commits changes the aspect of the histogram. It fixes a bug (nº 194 as mentionned in the commit message).

However, one thing that might be considered a drawback is that geom_histogram is now different from base::hist even when the breaks are the same:

d <- diamonds[diamonds$cut=="Fair",]
install_github("ggplot2", branch="1bfbee0fc791de8bfef901c156ebd04c4765b194")
ggplot(d, aes(carat)) + geom_histogram(binwidth = 0.2)
hist(d$carat, breaks=ggplot2:::fullseq(range(d$carat), size=0.2, pad=TRUE), freq=TRUE)
# -> same plot

ggplot(d, aes(carat, ..density..)) + geom_histogram(binwidth = 0.2)
hist(d$carat, breaks=ggplot2:::fullseq(range(d$carat), size=0.2, pad=TRUE), freq=FALSE)
# -> same plot

install_github("ggplot2", branch="bb4ec84bbedadd95647760fe586f64d0d6216abb")
ggplot(d, aes(carat)) + geom_histogram(binwidth = 0.2)
hist(d$carat, breaks=ggplot2:::fullseq(range(d$carat), size=0.2, pad=TRUE), freq=TRUE)
# -> different plot

Is this worth a "bug" report?

NB: after all this, if you don't want to be stuck in the past, redo
install.packages(c("plyr", "ggplot2"))
and start a new R session.

Jean-Olivier Irisson
---
Observatoire Océanologique
Station Zoologique, B.P. 28, Chemin du Lazaret
06230 Villefranche-sur-Mer
Tel: +33 04 93 76 38 04
Mob: +33 06 21 05 19 90
http://jo.irisson.com/

[Winston Chang]

Nice sleuthing! But doesn't the right=TRUE argument make it work the old way? (If you look at ?stat_bin, it describes 'right', though it would also help if there was an example in geom_histogram.)

ggplot(diamonds, aes(carat, ..density..)) + 

  geom_histogram(binwidth = 0.2, right = TRUE) +

  facet_grid(.~cut)

-Winston

[Jean-Olivier Irisson]

On 2012-Apr-06, at 09:40 , Winston Chang wrote:
>
> Nice sleuthing! But doesn't the right=TRUE argument make it work the old way? (If you look at ?stat_bin, it describes 'right', though it would also help if there was an example in geom_histogram.)

It does and the commit that I highlighted in the previous email basically changes right from TRUE to FALSE. What I am saying is that it might be better to default to what hist() does, to emphasize the fact that ggplot is a (better!) alternative to the regular plotting system of R.

If people get a different result for something as simple as an histogram, they might start to question the validity of gglot (even though both plots are mathematically valid and just use a different point of view). There was a similar debate a few months ago regarding qqnorm and stat_qq and it closed by basing the behaviour of stat_qq on qqnorm I think. And indeed the results are now the same:

set.seed(123)
x <- rnorm(15)
qqnorm(x)
library("ggplot2")
qplot(sample=x)

[Winston Chang]

It looks like there are slight differences when I compare ggplot2 0.8.9 to 0.9.0. (These are both run on R 2.14.2 on a Mac -- I'm not comparing 2.14 vs 2.15.)

I ran the following code with 0.8.9 and 0.9.0 (and I changed the filenames appropriately).

p <- ggplot(diamonds, aes(carat, ..count..)) + geom_histogram(binwidth = 0.2)

png('0.8.9-facet-grid.png', width=600, height=400)

p + facet_grid(. ~ cut)

dev.off()

png('0.8.9-facet-wrap.png', width=600, height=400)

p + facet_wrap( ~ cut, ncol=5)

dev.off()

I also tested against facet_wrap, since there is an existing bug with facet_grid where it drops duplicate data (), but that doesn't seem to be the issue here.

The results are attached. It looks like something has changed...

-Winston