alpha needs a scale and a legend even with discrete variables

[jiho]

Consider this use case, where I plot a series over rectangles which represent a binary response

d <- data.frame(x=1:50, y=runif(50), z=rep(c(T,F,T,T,F), each=10))
ggplot(d) + geom_rect(aes(xmin=x-0.5, xmax=x+0.5, ymin=-Inf, ymax=Inf, fill=z)) + geom_path(aes(x=x, y=y))

Rather than the color, I would prefer to have "something" when z is TRUE et nothing when z is FALSE (simpler, clearer, etc.). So:

ggplot(d) + geom_rect(aes(xmin=x-0.5, xmax=x+0.5, ymin=-Inf, ymax=Inf, alpha=z)) + geom_path(aes(x=x, y=y))

but then I have no way of knowing which is which because there is no legend... So I did:

ggplot(d) + geom_rect(aes(xmin=x-0.5, xmax=x+0.5, ymin=-Inf, ymax=Inf, fill=z, alpha=z)) + geom_path(aes(x=x, y=y))

But even then, the legend does not reflect the transparency of the FALSE parts.

[jiho]

This seems related to the fact that there is no scale_alpha_discrete, which is also a problem. Consider that, in the example above, I want the TRUE parts (those that show) to be half transparent so that the grid can still be seen underneath. I would do:

ggplot(d) + geom_rect(aes(xmin=x-0.5, xmax=x+0.5, ymin=-Inf, ymax=Inf, alpha=z)) + geom_path(aes(x=x, y=y)) + scale_alpha_continuous(to=c(0,0.5))

But this results in:

Error: Non-continuous variable supplied to scale_alpha_alpha_continuous.

First note the two "alpha" above. Then I should have been able to use scale_alpha_discrete in this case.

[jrandall]

The ggExtra package provides a scale_alpha_manual, which may be able to do what you want.

[hadley]

Works for me in devel.