LondonR 2014 - ggvis

By Matt Upson
25 November 2014, updated 7 April 2015

I went to my first LondonR meeting tonight hosted by Mango solutions. Some really great talks - especially presentatiosn by Matt Sundquist of plotly.

Mango solutions also presented a good introduction to ggvis and some of the interactive elements. I’ve included my notes from the event below. Note that the visualisations from ggvis will not render properly here. You will need to reproduce the document in RStudio to see them.

Note that much of the code used for ggvis had already become deprecated!

library(dplyr)
library(ggplot2)
 
tubeData <- read.table(
  "tubeData.csv",
  sep = ",",
  header = T
  )
 
str(tubeData)

## 'data.frame':	1050 obs. of  9 variables:
##  $ Line     : Factor w/ 10 levels "Bakerloo","Central",..: 1 1 1 1 1 1 1 1 1 1 ...
##  $ Month    : int  1 2 3 4 5 6 7 8 9 10 ...
##  $ Scheduled: num  29.4 29.4 29.3 29.3 29.3 ...
##  $ Excess   : num  6.04 6.54 4.77 5.4 5.23 5.03 5.14 5.73 4.8 5.95 ...
##  $ TOTAL    : num  35.5 36 34.1 34.7 34.5 ...
##  $ Opened   : int  1906 1906 1906 1906 1906 1906 1906 1906 1906 1906 ...
##  $ Length   : num  23.2 23.2 23.2 23.2 23.2 23.2 23.2 23.2 23.2 23.2 ...
##  $ Type     : Factor w/ 2 levels "DT","SS": 1 1 1 1 1 1 1 1 1 1 ...
##  $ Stations : int  25 25 25 25 25 25 25 25 25 25 ...

Outline

ggplot2
ggvis
%>%
Aesthetics
Layers
Interactivity

The Data

Tube performance data from TFL website.
Available here

ggplot2 recap

qplot or ggplot
Add layers with +
Change aesthetics by variable with aes
Control plot type with geom
Panel using facet_

head(tubeData)

##       Line Month Scheduled Excess TOTAL Opened Length Type Stations
## 1 Bakerloo     1     29.42   6.04 35.46   1906   23.2   DT       25
## 2 Bakerloo     2     29.42   6.54 35.96   1906   23.2   DT       25
## 3 Bakerloo     3     29.30   4.77 34.08   1906   23.2   DT       25
## 4 Bakerloo     4     29.30   5.40 34.70   1906   23.2   DT       25
## 5 Bakerloo     5     29.30   5.23 34.53   1906   23.2   DT       25
## 6 Bakerloo     6     29.30   5.03 34.33   1906   23.2   DT       25

qplot(
  data = tubeData,
  x = Month,
  y = Excess
  )

qplot(
  data = tubeData,
  x = Month,
  y = Excess,
  col = Line
  )

qplot(
  data = tubeData,
  x = Month,
  y = Excess,
  col = Line
  ) +
  facet_wrap(
    ~Line
    )

qplot(
  data = tubeData,
  x = Month,
  y = Excess,
  col = Line
  ) +
  facet_wrap(
    ~Line
    ) +
  geom_smooth(
    col = "red",
    size = 1
    )

The ‘geoms’

grep(
  "geom", 
  objects("package:ggplot2"), 
  value = TRUE
  )

##  [1] "geom_abline"          "geom_area"            "geom_bar"            
##  [4] "geom_bin2d"           "geom_blank"           "geom_boxplot"        
##  [7] "geom_contour"         "geom_crossbar"        "geom_density"        
## [10] "geom_density2d"       "geom_dotplot"         "geom_errorbar"       
## [13] "geom_errorbarh"       "geom_freqpoly"        "geom_hex"            
## [16] "geom_histogram"       "geom_hline"           "geom_jitter"         
## [19] "geom_line"            "geom_linerange"       "geom_map"            
## [22] "geom_path"            "geom_point"           "geom_pointrange"     
## [25] "geom_polygon"         "geom_quantile"        "geom_raster"         
## [28] "geom_rect"            "geom_ribbon"          "geom_rug"            
## [31] "geom_segment"         "geom_smooth"          "geom_step"           
## [34] "geom_text"            "geom_tile"            "geom_violin"         
## [37] "geom_vline"           "update_geom_defaults"

Facetting

Panels using facet_wrap and facet_grid.

Scales and themes

axes and styles
themes e.g. theme_bw etc

qplot(
  data = tubeData,
  x = Month,
  y = Excess,
  col = Line
  ) +
  facet_wrap(
    ~Line
    ) +
  geom_smooth(
    col = "red",
    size = 1
    ) +
  theme_bw()

Getting started with ggvis

Plot with ggvis function
Only a single function unlike ggplot1
Use ~ when referring to variables in a dataset, e.g. ~Ozone
This refers to variables as formulas
First variable always data.

require(ggvis)
 
myPlot <- ggvis(
  tubeData,
  ~Month,
  ~Excess
  )
 
 
# Creates a ggvis object:
 
class(myPlot)

## [1] "ggvis"

# Graphic is produced in the Viewer pane, not the Plots pane. Works via java vega a .d3 package
 
myPlot

# Note settings cog in the top right which allows you to change the rendering of teh plot.
 
# Can view in web browser and then be saved as an html file.
# Because it is not written to standard plotting device, you need to render the graphoc before you can save it out - i.e. no png or pdf command
# No equivalent script to save out of ggvis - must be saved from a browser
 
layer_points(myPlot)

# Can also be used in the pupe
 
myPlot %>% layer_points

The %>% operator

ggvis uses %>% from magrittr like dplyr

mean(airquality$Ozone,na.rm=TRUE)

## [1] 42.12931

# Now with the pipe
 
airquality$Ozone %>% mean(na.rm = TRUE)

## [1] 42.12931

# dplyr example
 
require(dplyr)
 
tubeData %>%
  dplyr::group_by(Line) %>%
  dplyr::summarise(mean = mean(Excess)) %>%
  qplot(Line, mean, data = ., geom="bar", stat = "identity", fill = Line)

%>% in ggvis

We pass ggvis objects mostly.
All functions accept a ggvis object first, except the command ggvis
Initial ggvis object is created with the ggvis command.
e.g.:

tubeData %>%
  ggvis(
    ~Month,
    ~Excess
    ) %>%
  layer_points

Changing properties

Properties in ggvis are the same as aesthetics in ggplot2
Number of aesthetics that can be set:
stroke – refers to lines
fill
size
opacity – instead of alpha

Changing based on variables

Mapping and setting as with aes
Map a variable to a property with =
Remember to use ~ with all variable names
fill = ~Line would set the fill based on the Line variable

tubeData %>%
  ggvis(
    ~Month,
    ~Excess
    ) %>%
  layer_points(
    fill = ~Line
    )

tubeData %>%
  ggvis(
    ~Month,
    ~Excess
    ) %>%
  layer_points(
    fill = ~Line,
    shape = ~Line
    )

tubeData %>%
  ggvis(
    ~Month,
    ~Excess
    ) %>%
  layer_points(
    size = ~Stations
    )

# can be set for all layers:
 
tubeData %>%
  ggvis(
    ~Month,
    ~Excess,
    fill = ~Line
    ) %>%
  layer_points

Setting property values

Instead of col = I("red") in ggplot2 is not required. This prevents ggplot2 picking red up as a fcator.
fill := "red" will work in ggvis

tubeData %>%
  ggvis(
    ~Month,
    ~Excess,
    fill = "red",
    opacity := 0.5
    ) %>%
  layer_points

tubeData %>%
  ggvis(
    ~Month,
    ~Excess,
    fill := "red",
    opacity := 0.5
    ) %>%
  layer_points

Shaping has changed in ggvis as it is dependent on .d3
At the moment a limited subset only is available

tubeData %>%
  ggvis(
    ~Month,
    ~Excess,
    fill := "red",
    opacity := 0.5,
    shape := "square"
    ) %>%
  layer_points

Exercise

Create a plot of mpg against wt using mtcars data
Use colour for the cyl variable, and make it a factor
Update the plotting symbol to be triangles

mtcars %>%
  ggvis(
    ~mpg,
    ~wt
    ) %>%
  layer_points(
    fill = ~factor(cyl),
    # Why doesn't this work!?
    shape := "triangle-up"
    )

Adding layers

In ggvis we use layer_ instead of geom_
Major limitation of ggvis at present, as not all of the geoms_ are vailable as layer_ in ggvis.
Check package manual:

tubeData %>%
  ggvis(
    ~Line,
    ~Excess
    ) %>%
  layer_boxplots()

# Adding some extra layers
 
mtcars %>%
  ggvis(
    ~mpg,
    ~wt
    ) %>%
  layer_points(
    fill = ~factor(cyl),
    # Why doesn't this work!?
    shape := "triangle"
    ) %>% 
  layer_smooths() %>%
  layer_model_predictions(
    model = "lm"
    )

# Note that formula can be specified with formula = ...
 
mtcars %>%
  ggvis(
    ~mpg,
    ~wt
    ) %>%
  layer_points(
    fill = ~factor(cyl),
    # Why doesn't this work!?
    shape := "triangle"
    ) %>% 
  layer_smooths(
    stroke := "blue",
    se = TRUE
    ) %>%
  layer_model_predictions(
    model = "lm",
    stroke := "red",
    se = TRUE
    )

Making plots interactive

Basic interactivity

Most basic level is ‘hover over’ just like in javascript.
Properties of the properties are changed to achive this.
property.hover argument: fill.hover := "red", or size.hover, opacity.hover, etc.

tubeData %>%
  ggvis(
    ~Month,
    ~Excess
    ) %>%
  layer_points(
    fill = ~Line,
    fill.hover := "red",
    size.hover := 1500 # sizes are very different to R graphics!
    )

# This behaviour is saved into the html or svg file!

Tooltips

add_tooltip adds other behaviour on hover..
We can provide a function that provide information as we hover.

tubeData %>%
  ggvis(
    ~Month,
    ~Excess
    ) %>%
  layer_points(
    fill = ~Line,
    fill.hover := "red",
    size.hover := 1500 # sizes are very different to R graphics!
    ) %>%
  add_tooltip(
    function(data) data$Excess
    )

# Locks off R console - cannot be used in markdown

pkData$id <- seq_along(pkData$Subject)
 
all_values <- function(x) {
  
  }
  
pkData %>% ggvis(
  ~Time,
  ~Conc,
  key = ~id # ggvis defined
  ) %>%
  layer_points() %>%
  add_tooltip(
    all_values,
    "hover"
    )

Interactive input

We can set outputs to be taken from interactive inputs

opacity := input_slider(0,1, label = "Opacity")

We use the ":=" for this input
We can optionally set labels next to the control - unlink shiny where it is not optional
Currently you are limited to changing the properties of the data, not the data itself.

tubeData %>%
  ggvis(
    ~Month,
    ~Excess
    ) %>%
  layer_points(
    fill = ~Line,
    size := input_slider(10,1000, label = "Size of points")
    )

Interactive input functions

tubeData %>%
  ggvis(
    ~Month,
    ~Excess
    ) %>%
  layer_points(
    size := input_numeric(30, label = "Size"),
    opacity := input_slider(0,1,value = 0.7, label = "Opacity"),
    fill := input_select(c("red","blue","orange"), label = "Colour")
    )

Common plot functions

Controlling axes and legends

We can control the axes using the add_axis function
This controls acis labels, tick marks and even grid lines
Title workaround is to use add_axis

add_axis("x", title = "Month")

add_axis controls colour of gridlines, etc
The add_legend and hide_legend functions allow use to control if we see a legend and wheere it appears

add_legend("fill") add_legend(c("fill","shape"))

Scales

ggvis had fewer scale functions than in ggplot2 but control much more.
just seven functions at present

grep(
  "^scale",
  objects("package:ggvis"),
  value = TRUE
  )

## [1] "scale_datetime" "scaled_value"   "scale_logical"  "scale_nominal" 
## [5] "scale_numeric"  "scale_ordinal"  "scale_singular"

ggvis vs ggplot2

we can layer graphics in a simlar fashion
aesthetics can be set baswed on by variables in the data
We cancontrol the type of plot

How are they different?

Only one main function
Layering with %>%
Fewer scale functions
Much functionality not available… but coming…

Which should I use

Static graphics: ggplot2
Interactive graphics ggvis

Documentation

ggvis documentation