Enchance subplot()'s ability to handle pre-specified domain(s)

[talgalili]

subplot seems the way to create a scatterplot with marginal histograms in plotly.

However, in order to have it work it needs to:

1. share the same axes so that all figures will respond to zoom together
2. we need control over each plots' width.

Example of how it should look like:

# Get some data
library(ggplot2 )
theme_set(theme_classic())
hist_top <- ggplot()+geom_histogram(aes(rnorm(100)))
empty <- ggplot()+geom_blank()
scatter <- ggplot()+geom_point(aes(rnorm(100), rnorm(100)))
hist_right <- ggplot()+geom_histogram(aes(rnorm(100)))+coord_flip()

library(gridExtra)
grid.arrange(hist_top, empty, scatter, hist_right, ncol=2, nrow=2, widths=c(4, 1), heights=c(1, 4))

Which looks like this:

And this is how it actually looks like (which should be corrected):

# hide axis ticks and grid lines
eaxis <- list(
  showticklabels = FALSE,
  showgrid = FALSE,
  zeroline = FALSE
)

p_empty <- plot_ly(filename="r-docs/dendrogram") %>%
  # note that margin applies to entire plot, so we can
  # add it here to make tick labels more readable
  layout(margin = list(l = 200),
         xaxis = eaxis,
         yaxis = eaxis)

subplot(hist_top, p_empty, scatter, hist_right, nrows=2, margin = 0.01)

[cpsievert]

The shared axes issue is reported in #298. The coord_flip() issue will be fixed in #390.

As for controlling the widths/heights of subplots -- for now you'll have to plotly_build() the subplot and alter the domains by hand:

x <- rnorm(100)
y <- rnorm(100)
p1 <- plot_ly(x = x, type = "histogram")
p2 <- plotly_empty()
p3 <- plot_ly(x = x, y = y, mode = "markers") 
p4 <- plot_ly(y = y, type = "histogram")
s <- subplot(p1, p2, p3, p4, nrows = 2)
l <- plotly_build(s)

# alter heights/widths
l$layout$xaxis$domain <- c(0, 0.8)
l$layout$xaxis2$domain <- c(0.8, 1)
l$layout$xaxis3$domain <- c(0, 0.8)
l$layout$xaxis4$domain <- c(0.8, 1)
l$layout$yaxis$domain <- c(0.8, 1)
l$layout$yaxis2$domain <- c(0.8, 1)
l$layout$yaxis3$domain <- c(0, 0.8)
l$layout$yaxis4$domain <- c(0, 0.8)
# axis titles/ticks should go too
l$layout$xaxis$title <- NULL
l$layout$xaxis$showticklabels <- F
l$layout$yaxis4$title <- NULL
l$layout$yaxis4$showticklabels <- F

It'd be nice if subplot() would respect layout(p, xaxis = list(domain = c(0, 0.8)) regardless of which axis it actually ends up being in the subplot, but I wouldn't consider that a high priority for now.

[talgalili]

Thanks @cpsievert - looking forward to updates on this.
Eventually being able to have this type of plot (scatterplot + marginal histograms), would be great.

With regards,
Tal

[talgalili]

Hi @cpsievert - I see here another way of creating this plot:
https://plot.ly/~snawel/80/brand-choices-for-age-60/
Do you have a suggestion in which way is better (or if there would be a new function in the near future that would make it easier?)

Thanks.

[royr2]

@talgalili - Here's something you could try. It (sort of) emulates your grid.arrange()example. Instead of dealing with subplot() you could use layout() and specify domains for each axis.

Hopefully this helps while @cpsievert is working on a more practical solution.

x <- rnorm(1000)
y <- rchisq(1000, df = 5)

ds <- data.frame(x,y)

ds %>% 
  plot_ly(x = x, y = y, type = "scatter", mode = "markers", showlegend = F, opacity = 0.7, 
          marker = list(symbol = "circle-dot",
                        size = 8)) %>%   # First trace
  add_trace(y = y, type = "histogram", xaxis = "x2", orientation = "h") %>%  # Second trace; Note the use of "x2"
  add_trace(x = x, type = "histogram", yaxis = "y2") %>%  # Second trace; Note the use of "x2"

  layout(title = "Marginal Histograms using Multiple Axis",
         xaxis = list(title = "X variable",
                      domain = c(0,0.85),
                      range = c(-3,3), 
                      showgrid = TRUE),
         xaxis2 = list(title = "Y Histogram",
                       domain = c(0.85,1), 
                       zeroline = FALSE),
         yaxis = list(title = "Y variable",
                      range = c(0,30), 
                      domain = c(0,0.85),
                      showgrid = TRUE),
         yaxis2 = list(title = "X Histogram",
                       domain = c(0.85,1),
                       zeroline = FALSE))

[talgalili]

Hi @royr2 - thank you for the code tip. I have a followup question:
I want to use add_trace and layout with the output of a ggplotly object. Sadly, in every variation I try I keep getting:

Error in FUN(X[[i]], ...) : 
  'options' must be a fully named list, or have no names (NULL)

Could you please show a similar example as you did, but using the output of the objects in my original question? (i.e.: hist_top, empty, scatter, hist_right)

Thanks!

Tal

[talgalili]

Code for reproducing the error:

x <- rnorm(1000)
y <- rchisq(1000, df = 5)
ds <- data.frame(x,y)

p <- ds %>% qplot(x = x, y = y, data = .)
py <- ds %>% ggplot(aes(x=x)) + geom_histogram(binwidth=.5) 

  ggplotly(p) %>%   # First trace
  # add_trace(y = y, type = "histogram", xaxis = "x2", orientation = "h") %>%  # Second trace; Note the use of "x2"
  add_trace(py, xaxis = "x2", orientation = "h") %>%  # Second trace; Note the use of "x2"
  add_trace(x = x, type = "histogram", yaxis = "y2") %>%  # Second trace; Note the use of "x2"

  layout(title = "Marginal Histograms using Multiple Axis",
         xaxis = list(title = "X variable",
                      domain = c(0,0.85),
                      range = c(-3,3), 
                      showgrid = TRUE),
         xaxis2 = list(title = "Y Histogram",
                       domain = c(0.85,1), 
                       zeroline = FALSE),
         yaxis = list(title = "Y variable",
                      range = c(0,30), 
                      domain = c(0,0.85),
                      showgrid = TRUE),
         yaxis2 = list(title = "X Histogram",
                       domain = c(0.85,1),
                       zeroline = FALSE))

Output is:

Error in FUN(X[[i]], ...) : 
  'options' must be a fully named list, or have no names (NULL)

[talgalili]

o.k., looking at this more, I see that my previous code wouldn't make sense. I see that add_trace takes a plotly object and adds another element to it (but it doesn't seem to allow the merging of two plotly objects).
Is there a way to merge two plotly objects into one layout?

If not then I guess the only way to get what I asked is (as the ticket says): "Enchance subplot()'s ability to handle pre-specified domain(s)"

This would be VERY useful for my work, so I hope this could get addressed.

Thanks,
Tal

[cpsievert]

There isn't any reason why subplot() couldn't support ggplot2 as well. Let's continue that discussion in #520

[cpsievert]

Fixed via dd9a3fa