Convert alpha aesthetics with ggplotly

Question

I have been working on heatmaps using in ggplot2, and I am now trying to convert them to ggplotly to build a shiny app with them. Everything works (more or less) fine except for the translation of the alpha aesthetic in geom_tile (same issue if I use geom_raster). After a full day of googling and reading related answers, I have not found any solution yet.

I have tried with different data and specifications, and so far no luck at all. The problem seems to be specific to heatmaps, as alpha translates well if I use geom_point. Problem is, I really need heatmaps for what I'm doing.

I have produced a reproducible example that shows the issue.

library(ggplot2)
library(plotly)
library(dplyr)

sample<- data.frame(a = 1:10,
                b = 11:20,
                c = 31:40,
                d = rep(c("a", "b"), 5))


plot <- sample%>%
  ggplot(aes(x=a, y=b, fill = c, alpha = d))+
  geom_raster()

plot

ggplotly(plot)

You'll see the output is quite different. Any ideas or has anyone had a similar issue?

ggplot:

ggplotly:

This is what I get: **Using alpha for a discrete variable is not advised.** Could you explain exactly what the issue is? I personally can't tell what the problem is since it's unclear what you expect to see. **but if you run it you'll see the output is quite different.** There's nothing to compare to. — NelsonGon, Jul 14 '19 at 12:37
Sorry, my fault, I had forgotten to print the plotly output in the sample. I just edited. If you run it now, you'll see that the ggplot and the plotly outputs are quite different (the latter does not seem to get the alpha). I have somehow found a way out of it by drawing the heatmap with geom_point(), but it's not really ideal... — aaumai, Jul 14 '19 at 14:57

score 1 · Answer 1 · answered Jul 16 '19 at 01:05

I agree there's something not being translated as intended by plotly::ggplotly(). I tried a few variations of ggplot2::geom_rect() & ggplot2::geom_tile() and ggplot2::scale_color_manual() & ggplot2::scale_color_identity().

Here's a heatmap example written straight to plotly, without the ggplot2 layer. I'm not seeing a directly way to map alpha/opacity.

This should meet your needs if you didn't need to vary the alpha in plotly, and simply needed the geom dimensions to be correct.

ds <- tibble::tibble(
  x   =  1:10,
  y   = 11:20,
  z   = 31:40,
  a1  = rep(c("a", "b"), 5),
  a2  = rep(c( .1,   .4), 5)
)

plot_ly(ds) %>% 
  add_heatmap(x = ~x, y = ~y, z = ~z, opacity = .5,  colors = "YlGnBu")

If you need to vary alpha, I'm guessing the best current approach is to create a custom palette that sets all four channels of rgba .

Thanks!! Yes, the issue is that I wanted to be able to vary the alpha, though this approach would have been great otherwise. I ended up creating a heatmap with geom_point which translated the alpha properly, I'll post it as an answer. — aaumai, Jul 18 '19 at 09:17

score 1 · Answer 2 · answered Jul 18 '19 at 09:22

In case anyone runs into a similar issue, I found a way around this (as it seems to be a bug of ggplotly) by creating the heatmap with geom_point() and adjusting the shape and size of the points until it looked the same. Apparently, ggplotly does translate well the alpha aes from this geom.

This is the final result (the aim of the shiny app is to be able to select these diagonals with a reference-year to isolate them from the others):

Convert alpha aesthetics with ggplotly

2 Answers2

Linked