Feature request: geom_ribbon: only fill area where ymax > ymin

[mkllr888]

In the aesthetics of geom_ribbon you specify a ymin and a ymax. However, geom_ribbon just fills the space between those two lines, without looking at minimum and maximum. (related to my question on stackexchange)

I propose that geom_point should only fill the area where ymax > ymin.

[jonocarroll]

As per my response on stackoverflow, this can be achieved using logic to assign fill within aes(.

## fill dependent on level > level2
h + 
  geom_ribbon(aes(ymin = level, ymax = level2, fill = level > level2)) +
  geom_line(aes(y = level)) + geom_line(aes(y = level2)) +
  scale_fill_manual(values = c("red", "green"), name = "fill")

Voting to close #1642 as not necessary.

The lack of interpolated fill where no data exists is either a feature or a bug, I'm not sure.

[sschmitz]

I vote to reopen this FR. The solution presented by @jonocarroll unfortunately does not work in the general case.

What happens is that the result of the expression assigned to the fill aesthetic is coerced to a factor, and thus, by default, contributes to the group aesthetic. geom_ribbon thus draws two ribbons, one for each factor level. Here, the expression is TRUE for all years up to 1923, and FALSE afterwards. This also explains the interpolated fill issue: the TRUE group ends at 1923, and so does the ribbon, meaning the green area is cut off at 1923. As such, the "lack of interpolated fill" is neither a bug nor a feature here, but simply expected behavior of the group aesthetic. (In the year 1924 (and from 1965 on), level happens to be exactly equal to level2, so the red ribbon correctly starts at the intersection since this is not interpolated, but an actual data point. This is also the reason why @jonocarroll's second SO solution works here, but also not in general.)

Let's use a crafted, less noisy example:

d <- data.frame(x = 0:5, y = c(1, 1, -1, -1, -1, -1))

ggplot(d, aes(x = x, y = y)) +
  geom_area(aes(fill = y > 0)) +
  geom_line() + geom_line(y = 0)

This is the same situation as in the above dataset. Note that the expression for the fill aesthetic starts at one of its two factor levels, and at some point switches to the other one. It does not, however, switch back. If it does:

d <- data.frame(x = 0:5, y = c(1, 1, -1, -1, 1, 1))

ggplot(d, aes(x = x, y = y)) +
  geom_area(aes(fill = y > 0)) +
  geom_line() + geom_line(y = 0)

Oops. Now both the left and the right side of the graph fall into the same group, while the middle part does not. The ribbon for the TRUE case, however, has no idea about the intermittent data, and just draws its ribbon across it.

From a user perspective, the easiest solution would probably be a fill2 (or something) aesthetic for geom_ribbon which is applied wherever ymax < ymin, and defaults to fill (leading to the current behavior unless specified). I guess the interpolation is already done somewhere inside the geom, since we do have interpolated fill in the current, single-color behavior. Another possibility would be a stat that interpolates the intersection and provides a grouping variable which basically is the same as ymax < ymin, but with the intersection repeated for both groups.

[jonocarroll]

As per my SO answer, "I don't think this is a ggplot2 bug, I think this is expected behaviour."

You mention my solution of using two ribbons "works here, but also not in general" - what is wrong in general where this doesn't work?

[sschmitz]

As I understood it, the two-ribbon solution is supposed to fix the non-interpolating fill issue, and does so by specifying level >= level2 for one ribbon, and level <= level2 for the other one. The edge case level == level2 is thus included in both ribbons. As I described, such a point exists in the SO dataset (in the year 1924) between the two regimes. My crafted data do not contain such a point, and thus, the approach fails:

d <- data.frame(x = 0:5, y = c(1, 1, -1, -1, -1, -1))

ggplot() +
  geom_ribbon(data = d[d$y >= 0, ], aes(x = x, ymin = 0, ymax = y), fill = "green") +
  geom_ribbon(data = d[d$y <= 0, ], aes(x = x, ymin = 0, ymax = y), fill = "red") +
  geom_line(data = d, aes(x = x, y = y)) + geom_line(data = d, aes(x = x), y = 0)

It also exhibits the same problems for the second example:

d <- data.frame(x = 0:5, y = c(1, 1, -1, -1, 1, 1))

ggplot() +
  geom_ribbon(data = d[d$y >= 0, ], aes(x = x, ymin = 0, ymax = y), fill = "green") +
  geom_ribbon(data = d[d$y <= 0, ], aes(x = x, ymin = 0, ymax = y), fill = "red") +
  geom_line(data = d, aes(x = x, y = y)) + geom_line(data = d, aes(x = x), y = 0)

[chamaoskurumi]

I second @sschmitz . I am also desperatly trying to find a solution for the case where there is no level == level2 edge observation in the data. Solving this issue would allow me to plot beautiful revenue vs. spending line graphs with a ribbon visualising the balance.

[hadley]

I think this is better fixed outside of ggplot2 - you need some way to generate split a single ribbon into multiple polygons. If someone can implement an elegant algorithm to do so, I'd consider adding to ggplot2.

[hadley]

That said, I'm surprised that these generate two different plots:

d <- data.frame(x = 0:5, y = c(1, 1, -1, -1, -1, -1))

ggplot(d, aes(x = x)) +
  geom_ribbon(aes(ymin = 0, ymax = y))

ggplot(d, aes(x = x)) +
  geom_area(aes(y = y))

[hadley]

Oh because geom_area() uses position_stack() which is going to cause problems here because we haven't cut up into multiple pieces (and position_stack() assumes bars, which don't have the same problem).