Google Analytics Data Explorer
Example 2: Google Analytics 2-D data explorer
This was Tim’s first web-based, interactive R applications. In hindsight…it was ambitious. It’s available at https://gilligan.shinyapps.io/ga_explorer_demo/.
Some keys to the underlying functionality used to build this:
- The
RGApackage – one of several packages available for pulling data from Google Analytics - The
shinypackage – Shiny is a package that makes R scripts interactive and web-based. - The
ggplot2package – the visualizations are all built using the plotting capabilities of ggplot2
The Code
This code does not use best practices for coding in general or for R in particular. But…it works. And, in this case, that’s what mattered. The typical way to build Shiny apps is to have two scripts: one defines the front end (ui.R) and the other defines the behind-the-scenes work (server.R). They reference each other, and it can be confusing as to how exactly that happens.
ui.R
library(shiny)
shinyUI(fluidPage(
# This feels a little hacky, but the default <h2> is just too big.
HTML('<style type="text/css">h2 {font-size: 150%;}</style>'),
# Application title
titlePanel(
"Traffic Explorer"),
fluidRow(
column(4,wellPanel(
# 'start' is hacked to default to the last full months; this needs to be cleaned up
dateRangeInput('dateRange',
label = 'Select date range:',
start = Sys.Date()-15, end = Sys.Date()-1
),
# Brand Selector
uiOutput("brandSelector"),
# X Dimension Selector
uiOutput("dimXSelector"),
# Y Dimension Selector
uiOutput("dimYSelector"),
# Date Granularity
uiOutput("granularitySelector"),
# Metric Selector
uiOutput("metricSelector")
)),
column(8,plotOutput("plotTotal"),
hr(),
plotOutput("plotTrends"))
)))server.R
library(shiny) # Package for making this thing run
library(RGA) # Package for getting GA data
library(ggplot2) # Package for rendering the graphics
library(dplyr) # Efficient manipulation of the data output.
library(scales) # To allow numbers to be formatted with commas and improved date formatting
# Authorize the Google Analytics account
ga_token <- authorize(client.id = "[insert Google Analytics API client ID]",
client.secret = "[insert Google Analytics API client secret]",
cache = "token")
#########################
# Configuration and loading of options
#########################
# List of sites to choose from. This includes the name for the site as well as
# the view ID for the site. Replace the values in brackets (and the brackets) with
# site names and corresponding view IDs. This list can be as long or as short as you
# want it to be.
brandMaster <- list("[Site 1 Display Name]"="[View ID]",
"[Site 2 Display Name]"="[View ID]",
"[Site 3 Display Name]"="[View ID]")
########################
# No additional edits are REQUIRED after this point in order for the app to work. You certainly
# CAN make additional edits, but only the values in brackets above must have updated valued.
########################
# Define the segment snippets. This is just a matrix where each row has the group name in the 1st column,
# the individual segment name in the second column, and the segment definition in the third column.
# The segment definitions use GA's "dynamic segment" syntax. NOTE: the "sessions::" or "users::"
# component of the segment definition is NOT included here because that gets added later. This
# script combines the snippets to make a segment for each "box" on the grid of results. Generally
# speaking, the segment snippets within any individual group should be mutually exclusive. Otherwise,
# the results will be tricky to interpret. Feel free to add additional groups/segments to this list!
segMaster <- matrix(c("Paid vs. Non-Paid","Paid Traffic",
"condition::ga:channelGrouping=~(Paid.Search)|(Display)|(Video)|(Social)|(Email)|(Other)"),1,3)
segMaster <- rbind(segMaster,c("Paid vs. Non-Paid","Non-Paid Traffic",
"condition::ga:channelGrouping=~(Organic.Search)|(Direct)|(Referral)"))
segMaster <- rbind(segMaster,c("Device Category","Desktop",
"condition::ga:deviceCategory==desktop"))
segMaster <- rbind(segMaster,c("Device Category","Tablet",
"condition::ga:deviceCategory==tablet"))
segMaster <- rbind(segMaster,c("Device Category","Mobile",
"condition::ga:deviceCategory==mobile"))
segMaster <- rbind(segMaster,c("New vs. Returning","New Visitors",
"condition::ga:userType==New Visitor"))
segMaster <- rbind(segMaster,c("New vs. Returning","Returning Visitors",
"condition::ga:userType==Returning Visitor"))
colnames(segMaster) <- c("group","segName","segDef")
# Set up the data granularity options
granularityMaster <- list("By Day"="ga:date",
"By Week"="ga:week",
"By Month"="ga:yearMonth")
# Set up the available metrics. With this version of the code, the only metrics that will be reliable
# are ones that are additive. In other words, ga:users will pull in data...but it will be misleading.
metricMaster <- list("Sessions"="ga:sessions",
"Page Views"="ga:pageviews",
"Bounces"="ga:bounces")
# Initialize a list that will hold the detailed data.
##### QUESTION: Is this needed? Is this where it should be? #####
allData <- data.frame(brand=character(0),viewId=character(0),segment.X=character(0),
segment.Y=character(0), date=character(0),metric=numeric(0))
# Define the base theme for the plots. Each plot will selectively override this stuff as needed
baseTheme <- theme(plot.title = element_text(face = "bold", size=16, colour="gray10"),
axis.line = element_blank(),
axis.text = element_text(face = "bold", size = 15, colour = "gray40"),
axis.ticks = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
panel.border = element_blank(),
panel.background = element_rect(fill="gray99"),
legend.position = "none",
strip.background = element_blank(),
strip.text = element_text(face = "bold", size = 15, colour = "gray40"))
shinyServer(function(input, output) {
######################################################################################
# Set up Input Selectors
######################################################################################
# Set up the list to allow the user to choose which brand/site to use
output$brandSelector <- renderUI ({
brands <- names(brandMaster) # Grab just the brands (not the view IDs)
defaultBrand <- brands[1] # Set the brand that is selected by default as the first one in the list
selectInput("brand", label = "Choose a site:",
choices = brands,
selected = defaultBrand)
})
# Set up the list to allow the user to choose what to use for the Y dimension
output$dimXSelector <- renderUI ({
dimOptions <- unique(segMaster[,1])
defaultDimX <- dimOptions[1] # Set the group that is selected by default as the first one in the list
selectInput("dimX", label = "Choose the X dimension:",
choices = dimOptions,
selected = defaultDimX)
})
# Set up the list to allow the user to choose what to use for the Y dimension
output$dimYSelector <- renderUI ({
dimOptions <- unique(segMaster[,1])
defaultDimY <- dimOptions[2] # Set the group that is selected by default as the first one in the list
selectInput("dimY", label = "Choose the Y dimension:",
choices = dimOptions,
selected = defaultDimY)
})
# Set up the list to allow the user to choose the data granularity for the trendlines
output$granularitySelector <- renderUI ({
granularityOptions <- names(granularityMaster) # Grab just the brands (not the view IDs)
# Set the granularity that is selected by default as the first one in the list
defaultGranularity <- granularityOptions[1]
selectInput("granularity", label = "Choose trendline resolution:",
choices = granularityOptions,
selected = defaultGranularity)
})
# Set up the list to allow the user to choose the metric to use for the results
output$metricSelector <- renderUI ({
metricOptions <- names(metricMaster) # Grab just the brands (not the view IDs)
defaultMetric <- metricOptions[1] # Set the metric that is selected by default as the 1st in the list
selectInput("metric", label = "Choose metric:",
choices = metricOptions,
selected = defaultMetric)
})
#############################
# Main function to pull the data based on selected inputs.
#############################
getDataFromGA <- reactive({
req(input$brand) ########## Do I need this if I have a default value set?
c.brandName <- input$brand
c.viewId <- brandMaster[[input$brand]]
c.metric <- metricMaster[[input$metric]]
c.granularity <- granularityMaster[[input$granularity]]
c.dimXlabel <- input$dimX
c.dimYlabel <- input$dimY
segmentNames.X <- segMaster[segMaster[,"group"]==c.dimXlabel,2]
segmentDefinitions.X <- segMaster[segMaster[,"group"]==c.dimXlabel,3]
segmentNames.Y <- segMaster[segMaster[,"group"]==c.dimYlabel,2]
segmentDefinitions.Y <- segMaster[segMaster[,"group"]==c.dimYlabel,3]
# Pull data for each segment. This is just a loop that goes through and makes successive calls to
# GA and then builds out the data frame with the data.
for (j in 1:length(segmentDefinitions.X)) {
# Get the segment details for the x-axis
c.segmentName.X <- segmentNames.X[j]
c.segmentDefinition.X <- segmentDefinitions.X[j]
for (i in 1:length(segmentDefinitions.Y)) {
# Get the segment details for the x-axis
c.segmentName.Y <- segmentNames.Y[i]
c.segmentDefinition.Y <- segmentDefinitions.Y[i]
# Combine the two segments. This is, basically, the segment to get "1 box" on the
# heatmap (and the trended detail)
c.segmentDefinition.XY <- paste(c("sessions::",c.segmentDefinition.X,";",
c.segmentDefinition.Y),collapse="")
# Get the users who visit from that segment -- by month
gaData <- get_ga(profileId = c.viewId,
start.date = input$dateRange[1], end.date = input$dateRange[2],
metrics = c.metric, dimensions = c.granularity, sort = NULL, filters = NULL,
segment = c.segmentDefinition.XY, samplingLevel = "HIGHER_PRECISION",
start.index = NULL, max.results = NULL, include.empty.rows = NULL,
fetch.by = NULL, ga_token)
# Combine the data with the meta data about it
tempData <- data.frame(brand = c.brandName, viewId=c.viewId,
segment.X=c.segmentName.X, segment.Y=c.segmentName.Y,
date = gaData[,1], metric = gaData[,2])
# Add the data to the data frame
allData <- rbind(allData,tempData)
}
}
# Reset the column names just to make sure they're right. The absence of this was causing
# the last column to not be named "metric" in some cases, which caused issues later in the code.
colnames(allData) <- c("brand", "viewId", "segment.X", "segment.Y", "date", "metric")
return(allData)
})
###############################
# Plot the data
###############################
# Make the heatmap plot
output$plotTotal <- renderPlot({
allData <- getDataFromGA()
# Collapse the values to remove the monthly breakdown
allData_total <- allData %>% group_by(segment.X, segment.Y) %>%
summarise(metric = sum(metric)) %>%
ungroup()
# We're going to display the % of total, too, so going ahead and calculating
# the total
metricTotal <- sum(allData_total$metric)
# Plot the result total
ggplot(allData_total,aes(x = segment.X, y = segment.Y, fill = metric)) +
geom_tile() + scale_fill_gradient(low="white", high="green") +
geom_text(aes(fill = allData_total$metric, label = comma(allData_total$metric)),
size=8, colour = "black", position = position_nudge(y = 0.1)) +
geom_text(aes(fill = allData_total$metric, label = percent(allData_total$metric/metricTotal)),
size=6, colour = "gray40", position = position_nudge(y = -0.1)) +
labs(title = paste(c("Total " ,input$metric," (",comma(sum(allData_total$metric)),")"),collapse="")) +
baseTheme + theme(panel.background = element_blank()) +
scale_x_discrete("") +
scale_y_discrete("")
})
# Plot trendlines for each block.
output$plotTrends <- renderPlot({
allData <- getDataFromGA()
# Reverse the order of the y-axis factors so the facets will appear in the
# same order as the heatmap totals above. This seemed like it was going to be
# unbelievably messy to do... but this does it.
allData$segment.Y <- factor(allData$segment.Y, levels=rev(levels(allData$segment.Y)))
ggplot(data = allData, mapping = aes(x = date, y = metric, group=1)) +
geom_line() +
facet_grid(segment.Y ~ segment.X, switch="y") +
baseTheme +
labs(title = paste(c(input$metric," Trended " ,input$granularity),collapse="")) +
scale_y_continuous(name="", labels = comma) +
theme(axis.title = element_blank(),axis.text.x = element_blank(),
strip.text.y = element_text(angle=180, hjust=1),
axis.text = element_text(face = "plain", size = 11, colour = "gray40"))
})
})