Lab 5 - Visualizations

Prerequisite Code & Setup

Setting global defaults for code chunks

knitr::opts_chunk$set(
  comment = '', fig.width = 6, fig.height = 6,
  warning = FALSE, error = FALSE, message = FALSE,
  include = TRUE, echo = TRUE, strip.white = TRUE,
  tidy = TRUE, highlight = TRUE
)

Loading libraries and data

libraries <- c("tidyverse","completejourney","lubridate","gridExtra","ggridges","naniar", "RColorBrewer", "hexbin")

# Load multiple packages
lapply(libraries, require, character.only = TRUE)    

# Full complete journey dataset... because slow loading is cool
promotions <- get_promotions()
transactions <- get_transactions()

dim(promotions)

dim(demographics)

dim(campaigns)

dim(campaign_descriptions)

Creating Visualizations

str(promotions)

str(demographics)

str(campaign_descriptions)

str(campaigns)

intersect(colnames(campaigns),colnames(demographics))

intersect(colnames(campaign_descriptions),colnames(campaigns))

data_clean <- demographics %>% 
  inner_join(campaigns, by = "household_id") %>% 
  inner_join(campaign_descriptions, by = "campaign_id") %>%
  mutate(camp_length = as.duration(end_date - start_date))

# Range of dates, number of NA's
data_clean %>% 
  summarize(max_date = max(end_date),
            min_date = min(start_date),
            nas = sum(is.na(data_clean)))

p1 <- data_clean %>%
  group_by(income) %>% 
  summarize( mean_days = (mean(camp_length/86400))) %>% 
  arrange(desc(mean_days)) %>% 
  ggplot(aes(x = reorder(income, -mean_days), y = mean_days, 
             fill = reorder(income, -mean_days))) +
    geom_bar(stat = "identity", show.legend = FALSE) +
    scale_y_continuous() +
    scale_fill_grey() +
    labs(title = "Mean Length (Days) of Marketing Campaigns by Income Group",
         subtitle = 
        "Households with income under 15k see the longest avg. campaign times.",
         caption = "Based off the CompleteJourney data set by Brad Boehmke",
        x = "Household Income Group",
        y = "Avg. Length of Campaign"
    )

p2 <- data_clean %>% 
  filter(income == "Under 15K") %>% 
  group_by(campaign_type) %>%
  ggplot(aes(x = campaign_type, y = (camp_length)/86400, 
             fill = campaign_type)) +
    geom_violin() +
    scale_fill_brewer(type = "qual", palette = "Set2",
                      name = "Campaign Type") +
  geom_jitter(stat = "identity", width = 0.07, 
              color = "black", alpha = 0.25) +
  labs(title = "Campaign Length By Campaign Type for Houses with Income Less Than $15k",
       subtitle = "Drilling down into campaigns for houses with income less than 15K, campaign type C has the longest typical campaign time hovering \n between 60-80 days per campaign. Additionally, campaigns A and B occur more often than Campaign C (higher frequency), but \n they have lower typical durations in days",
       x = "Campaign Type",
       y = "Length of Campaign (days)")

data_bigger <- data_clean %>% 
  inner_join(transactions, by = "household_id") %>% 
  inner_join(products, by = "product_id") %>% 
  filter(campaign_type == "Type C") %>% 
  select(product_id, product_type, quantity, camp_length) %>% 
  group_by(product_id, product_type) %>% 
  summarize(med_qty = median(quantity, na.rm = TRUE),
            med_length = median(as.numeric(camp_length/86400), na.rm = TRUE)) %>%
  filter(med_qty > 0) %>% 
  arrange(desc(med_qty))

qty_mn <- mean(data_bigger$med_qty, na.rm = TRUE)
qty_sd <- sd(data_bigger$med_qty, na.rm = TRUE)

LOT_mn <- mean(data_bigger$med_length, na.rm = TRUE)
LOT_sd <- sd(data_bigger$med_length, na.rm = TRUE)

p3 <- data_bigger %>% 
  filter(med_qty <= qty_mn + 3*qty_sd & med_qty >= qty_mn - 3*qty_sd) %>% 
  group_by(product_id) %>% 
  arrange(desc(med_qty)) %>% 
  ggplot(aes(x = med_length, y = med_qty)) + 
   geom_hex(bins = 15, show.legend = FALSE) +
   viridis::scale_fill_viridis(option = "E", alpha = 0.8) +
  labs(title = "Median Quantity Sold Compared to Median Length of a Campaign Type C by Product_id",
       subtitle = "Most C-Type Campaigns run for 50-60 days, with the typical product that campaign C applies to being sold in \n quantities of 1-5 units per transaction. In other words, Campaign C often runs for 50-60 days on low-quantity items.",
       x = "Median Length of Campaign Time (days) by Product",
       y = "Median Quantity Sold per Transaction by Product",
  )

Visualization Outputs

p1

p2

p3