Making an animation of the Bernoulli distribution using gganimate and getting an unexpected jump at p = 0.5

Question

I'm practicing using gganimate and want to create something similar to this Shiny app but animated. I've had some success, but am getting a really weird error that I honestly cannot explain and was hoping someone could help me out. The rest of this post contains a fully reproducible example.

Here is what I have so far:

bernoulli <- cbind.data.frame(c(rep(1,101), rep(0,101)), c(seq(1,0,by=-0.01), seq(0,1,by=0.01)), c(seq(1,0,by=-0.01), seq(1,0,by=-0.01)))
names(bernoulli) <- c("success", "probability", "p")
bernoulli <- bernoulli[order(bernoulli$p),]
row.names(bernoulli) <- c(1:nrow(bernoulli))

This creates the data frame I am working with. It's three variables, so very straightforward. The logic of my animation is to how how the probability of the dichotomous outcomes vary by the p parameter of the Bernoulli distribution. To give an example of a static graph, if I do something like this:

ggplot(subset(bernoulli, p == 0.70), aes(x=success, y=probability)) +
  geom_bar(stat="identity")

and vary the p by which I am subsetting on, I get the desired outcome (although for some reason, there are certain numbers, seemingly completely randomly, where I cannot subset the data, even though I can see them clearly in the data frame. I have checked that the variables are numeric and whatnot, so that's not the issue. This is a side problem that I have no idea why it is occurring).

For example, switch between p == 0.12 and p == 0.70 and compare them to the Shiny app mentioned earlier and you'll see it matches up, though the axes vary, of course.

When I try to implement the animation as follows:

ggplot(bernoulli, aes(x=success, y=probability)) +
  geom_bar(stat="identity") +
  transition_time(p)

Everything looks great! The axes are matched up nicely, the animation flows smoothly... but as soon as the bars hit p == 0.50, they both jump up to 1, something which is not reflected in the data at all. Is this just an issue with my computer/R/ggplot/gganimate?

score 3 · Accepted Answer · answered Jul 23 '19 at 23:54

3

You need to fix y axis using scale_y_continuous()

library(dplyr)
bernoulli %>%
    dplyr::filter(probability > 0 & probability < 1) %>%
    ggplot(aes(x=success, y=probability)) +
        geom_col() + 
        scale_y_continuous(limits = c(0,1)) +
        transition_time(p)

If you don't fix y axis, it will look like this when p==0.5:

And it is truly strange that if I don't do the filter above, one of the bar will still jump to 1 when p==0.5

answered Jul 23 '19 at 23:54

yusuzech

5,896
1
18
33

Thanks, that works. But it is still so odd that you need to do that filter to make it work. In fact, you can remove `scale_y_continous` and it will produce the exact same animation, but remove the filer (and keep the scale, or not) and you get that error. I'm guessing something is just off when the probability values are exactly 0 and exactly 1. – Julian Jul 25 '19 at 00:15

Making an animation of the Bernoulli distribution using gganimate and getting an unexpected jump at p = 0.5

1 Answers1