Ordering of traces is wrong

[jpevitt]

Hey guys, I'm pretty new to the plotly API so if I'm missing something please let me know. I have a 100% stacked area chart in ggplot2 that I'm trying to convert into a plotly plot. I followed the - pretty simple - steps on your API site which took me to the URL for it, except on the site my colors and/or lines entirely disappear leaving me with the entire plot blacked out in some random color.

[13bzhang]

Sorry to hear that you are experiencing problems. Do you have code and images that you can share with us?

[13bzhang]

@jpevitt Thanks for your reply. But I cannot find the attached folder. All I see if a a bunch of numbers. Can you post the code directly to this issue or upload your folder as a repo or a gist on GitHub?

[jpevitt]

Sorry, answered through email, didn't realize it wouldn't display properly.
Here is my code:
df <- read.delim(file.choose(), header=T)
dfm <- melt(df, id.vars="Year")
full <-
ggplot(data=dfm, aes(x=Year, y=value)) +
fte_theme(SCALE, MARGIN) +
geom_area(aes(fill=variable)) +
labs(title="Generic Title", x=NULL, y=NULL) +
theme(legend.position="right") +
theme(legend.title=element_blank()) +
scale_y_continuous(labels=percent)
library(plotly)
py <- plotly()
py$ggplotly(full)

It results in this ggplot plot when I render it in R

But on plotly it results in this:

[13bzhang]

Thanks for sharing. I will look into this.

[cpsievert]

I'm not exactly sure what's going wrong yet, but here is a reproducible example to help debug:

library(dplyr)
df <- diamonds %>% group_by(carat, cut) %>% 
  summarise(n = n()) %>% mutate(freq = n/sum(n))
p <- ggplot(data = df, aes(x = carat, y = freq, fill = cut)) + 
  geom_area()

[cpsievert]

Ah, this is related to the order in which the traces are drawn. If you hover over the example I provide above, you'll see hover info for all the traces. @13bzhang does that make sense? Feel free to tackle this one if you'd like.

[13bzhang]

Yeah I can take this one.

[13bzhang]

I have been working on this one for the past few days. So I noticed that stacked densities also don't work with the current code. For instance:

p <- ggplot(data = movies, aes(x = rating, fill = mpaa,)) + 
  geom_density(position = "stack")
py$ggplotly(p)

ggplot2

Plotly

The problem is that the order of the traces are laid in reverse. I played around with the code in trace_generation.R a bit and added the following:

   if (g$geom == "area" | g$geom == "density"){
     traces <- rev(traces)
   }

Again, this is really unsophisticated but just a first cut.

I was able to produce the following:

ggplot2

Plotly

Plotly

Note some problems still left:

• the legend is reversed in the area example
• the legend doesn't show up in the density example

Maybe the solution is to make special cases for these types of geoms like in
https://github.com/ropensci/plotly/blob/master/R/trace_generation.R#L251-L263
?

@cpsievert @chriddyp

[cpsievert]

It looks like this problem isn't necessarily specific to a geom. For example,

qplot(wt, drat, color = gear, data = mtcars)

leads to

That legend should have 3 at the top and 1 at the bottom. I think you can fix this by having trace "3" be the first trace, trace "2" be the second, and trace "1" be the third.

You may also need to add some special handling for the area geom, but I'm not sure.

[cpsievert]

@13bzhang sorry, I didn't read your post carefully. I wouldn't worry about the ordering of the legend, in fact, I think the plotly version is more sensible than ggplot2. Also, the legend hiding problem is reported in #212.

That being said, I'd start a PR with your ordering fix and include a test!

[13bzhang]

@cpsievert If it's a more general problem, then the fix should be somewhat easier.

[cpsievert]

Fixed via a4a9a1d