Dave Childers

• Data Scientist with Powerley
• Before that: Data Science Consulting
• Before that: CSCAR at UM

## Warning: Can't output dynamic/interactive ggvis plots in a knitr document.
## Generating a static (non-dynamic, non-interactive) version of the plot.

• Graphics moving to web
• Interactivity
• Fast Exploratory Data Analysis

1. Dependencies
2. Syntax
3. Exploratory Data Analysis
4. Interactivity

# easy to read/write
cocaine %>%
  mutate(log_weight = log(weight)) %>%
  filter(weight <= 10)

# hard to read/write
filter(mutate(cocaine, log_weight = log(weight)), weight <= 10)

pipe introduction

ggplot(cocaine, aes(x = weight, y = price)) +
  geom_point()

ggvis(cocaine, x = ~weight, y = ~price) %>%
  layer_points

ggplot(cocaine, aes(x = weight, y = price)) +
  geom_point()

ggvis(cocaine, x = ~weight, y = ~price) %>%
  layer_points

Similarities

• Define graphic by composing small functions
• Mapping variables to visual properties
• Property Inheritance

ggplot(cocaine, aes(x = weight, y = price)) +
  geom_point()

ggvis(cocaine, x = ~weight, y = ~price) %>%
  layer_points

Differences (ggplot -> ggvis)

1. plus -> pipe
2. layer -> geom
3. point -> points
4. aes() -> ~

# mapping state -> color
ggplot(cocaine, aes(x = weight, y = price)) +
  geom_point(aes(color = state))

# setting color to a constant value
ggplot(cocaine, aes(x = weight, y = price)) +
  geom_point(color = "orange")

ggvis(cocaine, x = ~price) %>% 
  layer_histograms(width = 100, center = 50)

ggvis(cocaine, x = ~price) %>% layer_densities %>% 
  add_axis("y", title_offset = 50)

ggvis(cocaine, x = ~state) %>% layer_bars

What happens if we say text = ~n ?

cocaine %>% count(state, sort = TRUE) %>%
  ggvis(x = ~reorder(state, -n), y = ~n) %>%
  filter(n >= 100) %>%
  add_axis("x", title = "State") %>%
  layer_bars %>%
  layer_text(text := ~n, fontSize := 20) 
  # what happens if we say text = ~n ?

cocaine %>% count(state, sort = TRUE) %>%
  ggvis(x = ~reorder(state, -n), y = ~n) %>%
  filter(n >= 100) %>%
  layer_bars %>%
  layer_text(text := ~n) %>%
  add_axis("x", title = "State") %>%
  add_axis(
    "x", 
    title = "2007 Cocaine Seizures by State", 
    orient = "top",
    ticks = 0,
    properties = axis_props(
      axis = list(stroke = "white"),
      labels = list(fontSize = 0)
    )
    )

• ggtitle()
• coord_flip()
• themes
• Faceting

%>% layer_smooths

%>% layer_model_predictions(model = "lm", formula = price ~ weight)

ggvis(cocaine, ~weight, ~price, fill = ~state, stroke = ~state) %>%
  filter(state %in% c("FL", "IN", "NY")) %>%
  mutate(log_weight = log(weight), log_price = log(price)) %>%
  layer_points(~log_weight, ~log_price, opacity := 0.2) %>%
  auto_group() %>%
  layer_model_predictions(
    model = "lm", 
    formula = log_price ~ log_weight
  ) %>%
  add_axis("x", title = "Log Price") %>%
  add_axis("y", title = "Log Weight")

methods(class = "ggvis")

##  [1] arrange_                 compute_align           
##  [3] compute_bin              compute_boxplot         
##  [5] compute_count            compute_density         
##  [7] compute_model_prediction compute_stack           
##  [9] compute_tabulate         distinct_               
## [11] explain                  filter_                 
## [13] group_by_                groups                  
## [15] knit_print               mutate_                 
## [17] print                    rename_                 
## [19] select_                  slice_                  
## [21] summarise_               transmute_              
## [23] ungroup                 
## see '?methods' for accessing help and source code

## Warning: Can't output dynamic/interactive ggvis plots in a knitr document.
## Generating a static (non-dynamic, non-interactive) version of the plot.

%>% layer_points(
    size := input_slider(min = 50, max = 500, value = 50, step = 50)
)

## Warning: Can't output dynamic/interactive ggvis plots in a knitr document.
## Generating a static (non-dynamic, non-interactive) version of the plot.

%>% add_tooltip(function(x) x$id) 

• cannot switch between data sets

• cannot add/remove layers

• need Shiny for full interactivity

• More interactive than ggplot

• Faster exploratory analysis than shiny

• Be cautious about using in production

• ggvis webpage
• Stackoverflow
• RStudio Blog

Contact

• Dave Childers
• childers.dave@gmail.com