How to set legend alpha with ggplot2

I have a graph of wind speeds against direction which has a huge numeber of points, and so am using alpha=I(1/20) in addition to color=month

Here is a sample of code:

library(RMySQL)
library(ggplot2)
con <- dbConnect(...)
wind <- dbGetQuery(con, "SELECT speed_w/speed_e AS ratio, dir_58 as dir, MONTHNAME(timestamp) AS month, ROUND((speed_w+speed_e)/2) AS speed FROM tablename;");

png("ratio-by-speed.png",height=400,width=1200)
qplot(wind$dir,wind$ratio,ylim=c(0.5,1.5),xlim=c(0,360),color=wind$month,alpha=I(1/30),main="West/East against direction")
dev.off()

This produces a decent graph, however my issue is that the alpha of the legend is 1/30th also, which makes it unreadable. Is there a way I can force the legend to be 1 alpha instead?

Here is an example:

Update With the release of version 0.9.0, one can now override aesthetic values in the legend using override.aes in the guides function. So if you add something like this to your plot:

+ guides(colour = guide_legend(override.aes = list(alpha = 1)))

that should do it.

---

I've gotten around this by doing a duplicate call to the geom using an empty subset of the data and using the legend from that call. Unfortunately, it doesn't work if the data frame is actually empty (e.g. as you'd get from subset(diamonds,FALSE)) since ggplot2 seems to treat this case the same as it treats NULL in place of a data frame. But we can get the same effect by taking a subset with only one row and setting it to NaN on one of the plot dimensions, which will prevent it from getting plotted.

Based off Chase's example:

# Alpha parameter washes out legend:
gp <- ggplot() + geom_point(data=diamonds, aes(depth, price, colour=clarity), alpha=0.1)
print(gp)

# Full color legend:
dummyData <- diamonds[1, ]
dummyData$price <- NaN
#dummyData <- subset(diamonds, FALSE)   # this would be nicer but it doesn't work!
gp <- ggplot() +
  geom_point(data=diamonds, aes(depth, price, colour=clarity), alpha=0.1, legend=FALSE) + 
  geom_point(data=dummyData, aes(depth, price, colour=clarity), alpha=1.0, na.rm=TRUE)
print(gp)

A bit of googling turned up this post which doesn't seem to indicate that ggplot currently supports this option. Others have addressed related problems by using gridExtra and using viewPorts as discussed here.

I'm not that sophisticated, but here's one approach that should give you the desired results. The approach is to plot the geom twice, once without an alpha parameter and outside of the real plotting area. The second geom will include the alpha parameter and suppress the legend. We will then specify the plotting region with xlim and ylim. Given that you are a lot of points, this will roughly double the plotting time, but should give you the effect you are after.

Using the diamonds dataset:

#Alpha parameter washes out legend
ggplot(data = diamonds, aes(depth, price, colour = clarity)) + 
geom_point(alpha = 1/10)

#Fully colored legend
ggplot() +
geom_point(data = diamonds, aes(depth, price, colour =clarity), alpha = 1/10, legend = FALSE) + 
geom_point(data = diamonds, aes(x = depth - 999999, y = price - 999999, colour = clarity)) +
xlim(40, 80) + ylim(0, 20000)