12

I have 5 clusters of x,y data I'm plotting using R's plotly.

Here are the data:

set.seed(1)
df <- do.call(rbind,lapply(seq(1,20,4),function(i) data.frame(x=rnorm(50,mean=i,sd=1),y=rnorm(50,mean=i,sd=1),cluster=i)))

Here's their plotly scatter plot:

library(plotly)
clusters.plot <- plot_ly(marker=list(size=10),type='scatter',mode="markers",x=~df$x,y=~df$y,color=~df$cluster,data=df) %>% hide_colorbar() %>% layout(xaxis=list(title="X",zeroline=F),yaxis=list(title="Y",zeroline=F))

Which gives: enter image description here

Then, following @Marco Sandri's answer, I add polygons circumscribing these clusters using this code:

Polygons code:

library(data.table)
library(grDevices)

splinesPolygon <- function(xy,vertices,k=3, ...)
{
  # Assert: xy is an n by 2 matrix with n >= k.
  # Wrap k vertices around each end.
  n <- dim(xy)[1]
  if (k >= 1) {
    data <- rbind(xy[(n-k+1):n,], xy, xy[1:k, ])
  } else {
    data <- xy
  }
  # Spline the x and y coordinates.
  data.spline <- spline(1:(n+2*k), data[,1], n=vertices, ...)
  x <- data.spline$x
  x1 <- data.spline$y
  x2 <- spline(1:(n+2*k), data[,2], n=vertices, ...)$y
  # Retain only the middle part.
  cbind(x1, x2)[k < x & x <= n+k, ]
}

clustersPolygon <- function(df)
{
  dt <- data.table::data.table(df)
  hull <- dt[,.SD[chull(x,y)]]
  spline.hull <- splinesPolygon(cbind(hull$x,hull$y),100)
  return(data.frame(x=spline.hull[,1],y=spline.hull[,2],stringsAsFactors=F))
}

library(dplyr)
polygons.df <- do.call(rbind,lapply(unique(df$cluster),function(l)
  clustersPolygon(df=dplyr::filter(df,cluster == l)) %>%
    dplyr::rename(polygon.x=x,polygon.y=y) %>%
    dplyr::mutate(cluster=l)))

And now adding the polygons:

clusters <- unique(df$cluster)

for(l in clusters) clusters.plot <- clusters.plot %>% 
 add_polygons(x=dplyr::filter(polygons.df,cluster == l)$polygon.x,
              y=dplyr::filter(polygons.df,cluster == l)$polygon.y,
              line=list(width=2,color="black"),
              fillcolor='transparent', inherit = FALSE)

Which gives:

enter image description here

Although this works great, unfortunately it eliminates the hoverinfo that existed prior to adding the polygons, and now is just the trace of each polygon.

Changing inherit from FALSE to TRUE results with the error I write about in that post. So my question is how to add the polygons without changing the hoverinfo of the original plot.

dan
  • 6,048
  • 10
  • 57
  • 125
  • The polygons hide the information underneath. Maybe you could replot the markers: clusters.plot %>% add_markers(x=~df$x,y=~df$y, showlegend = FALSE) – MLavoie Jan 05 '18 at 23:30
  • The hoverinfo is restored but now in addition to the polygons all points are connected by lines. – dan Jan 06 '18 at 01:01
  • Try setting `hoverinfo="none"` to the `add_polygon` call – alan ocallaghan Jan 06 '18 at 01:23
  • That just eliminated the "trace #of cluster" hoverinfo but didn't recover the hoverinfo of the points. – dan Jan 06 '18 at 01:30
  • Why not plotting the polygone before the markers? Also `polygons.df` is not properly defined, you missed some code lines there. – drmariod Feb 07 '18 at 06:42
  • And, have you tried `fillcolor='none'`? – drmariod Feb 07 '18 at 06:49
  • Can you edit your post to include the full contents of the line `polygons.df <- do.call(rbind,lapply(unique(df$cluster),function(l) `? Not quite able to reproduce `polygons.df` here. – Matt Summersgill Feb 12 '18 at 15:55

3 Answers3

7

I think part of the issue here is that the colorbar in plotly has some somewhat weird behavior and side effects when you start to mix and match trace types.

The simplest way to work around this (and it seems appropriate since you are coloring by clusters, not a continuous variable) is to change the class of your clustered column to be an ordered factor with the expression df$cluster <- ordered(as.factor(df$cluster)). (I believe this could be in a dplyr mutate statement as well.)

Packages and data generation functions

library(data.table)
library(grDevices)
library(dplyr)
library(plotly)

## Function Definitions 
splinesPolygon <- function(xy,vertices,k=3, ...) {
  # Assert: xy is an n by 2 matrix with n >= k.
  # Wrap k vertices around each end.
  n <- dim(xy)[1]
  if (k >= 1) {
    data <- rbind(xy[(n-k+1):n,], xy, xy[1:k, ])
  } else {
    data <- xy
  }
  # Spline the x and y coordinates.
  data.spline <- spline(1:(n+2*k), data[,1], n=vertices, ...)
  x <- data.spline$x
  x1 <- data.spline$y
  x2 <- spline(1:(n+2*k), data[,2], n=vertices, ...)$y
  # Retain only the middle part.
  cbind(x1, x2)[k < x & x <= n+k, ]
}

clustersPolygon <- function(df) {
  dt <- data.table::data.table(df)
  hull <- dt[,.SD[chull(x,y)]]
  spline.hull <- splinesPolygon(cbind(hull$x,hull$y),100)
  return(data.frame(x=spline.hull[,1],y=spline.hull[,2],stringsAsFactors=F))
}

Generate Data

The one critical difference here is to define your cluster as an ordered factor to keep it from being treated as a continuous variable that will invoke the colorbar weirdness.

set.seed(1)
df <- do.call(rbind,lapply(seq(1,20,4),function(i) data.frame(x=rnorm(50,mean=i,sd=1),y=rnorm(50,mean=i,sd=1),cluster=i)))

## Critical Step here: Make cluster an ordered factor so it will
## be plotted with the sequential viridis scale, but will not 
## be treated as a continuous spectrum that gets the colorbar involved
df$cluster <- ordered(as.factor(df$cluster))

## Make hull polygons
polygons.df <- do.call(rbind,lapply(unique(df$cluster),function(l) clustersPolygon(df=dplyr::filter(df,cluster == l)) %>% dplyr::rename(polygon.x=x,polygon.y=y) %>% dplyr::mutate(cluster=l)))
clusters <- unique(df$cluster)
clustersPolygon(df=dplyr::filter(df,cluster == l)) %>% dplyr::rename(polygon.x=x,polygon.y=y) %>% dplyr::mutate(cluster=l)))

Build a plotly object

Mostly the same here, but starting by initializing an empty plotly object and then adding the hull polygons before the raw data points.

## Initialize an empty plotly object so that the hulls can be added first
clusters.plot <- plot_ly()

## Add hull polygons sequentially
for(l in clusters) clusters.plot <- clusters.plot %>% 
  add_polygons(x=dplyr::filter(polygons.df,cluster == l)$polygon.x,
               y=dplyr::filter(polygons.df,cluster == l)$polygon.y,
               name = paste0("Cluster ",l),
               line=list(width=2,color="black"),
               fillcolor='transparent', 
               hoverinfo = "none",
               showlegend = FALSE,
               inherit = FALSE)  

## Add the raw data trace
clusters.plot <- clusters.plot %>% 
  add_trace(data=df, x= ~x,y= ~y,color= ~cluster,
            type='scatter',mode="markers",
            marker=list(size=10)) %>% 
  layout(xaxis=list(title="X",
                    zeroline=F),
         yaxis=list(title="Y",
                    zeroline=F))
## Print the output
clusters.plot

Gives the following output

polygons

Matt Summersgill
  • 4,054
  • 18
  • 47
2

This seems to give what you are looking for :

for(l in clusters) clusters.plot <- clusters.plot %>% 
  add_polygons(x=dplyr::filter(polygons.df,cluster == l)$polygon.x,
           y=dplyr::filter(polygons.df,cluster == l)$polygon.y,
           line=list(width=2,color="black"),type = "contour",
           fillcolor='transparent', inherit = FALSE)

I add the 

type = "contour"

Not sure the fillcolor is needed anymore .. Does it fit your need ?

MrSmithGoesToWashington
  • 1,048
  • 10
  • 23
2

A bit work around. The poly.df file can be replaced by your data.frame. One can simply ggplot for visualisation and then transform by ggplotly.

library(tidyverse)
library(plotly)

set.seed(1)
df <- do.call(rbind,lapply(seq(1,20,4), 
                           function(i) data.frame(x=rnorm(50,mean=i,sd=1),y=rnorm(50,mean=i,sd=1),cluster=i)))
poly.df <- df %>% 
  group_by(cluster) %>%
  do(.[chull(.$x, .$y),]) 

ggplot(df, aes(x, y, colour = as.factor(cluster))) +
  geom_polygon(data = poly.df, fill = NA)+
  geom_point() ->
  p

ggplotly(p)

enter image description here

Volodymyr
  • 888
  • 10
  • 21