1 Overview
In this take-home exercise, we will explore creating data visualisations for multidimensional data, using starbucks drinks and their nutritional value.
Source: Starbucks
1.1 Challenges Faced
The dataset had to be cleaned before usage as the caffeine field contains strings due to range of values given for some drinks. For simplicity, max value of the range was used to conduct a more conservative analysis. This was resolved using a for loop and ifelse() condition. The field was converted to a numeric datatype before analysis.
The dataset contained drinks with different combinations of milk, whipped cream. Furthermore, the nutritional values are based on different size and volume for each drink. For an objective analysis, the milk type and whipped cream had to be taken into consideration and normalised by unit volume before clustering and plotting the heatmap.
As this is my first time exploring in-depth into multidimensional data visualisations in R, it took me some time to get used to the customisation of the plot as compared to using ggplot2. By referring to the respective package documentations, I was able to adjust the size and colour of the correlation number label , glyphs, text labels on axis and colour labels on the corrplots.
2 Installing Packages
The following packages and libraries were installed for this exercise:
tidyverse : A collection of core packages designed for data science, used extensively for data preparation and wrangling.
knitr: Package used for dynamic report generation
rmarkdown: Used to convert R Markdown documents into a variety of formats.
corrplot: Used for plotting a correlation matrix, test for correlation, and other visualization methods about association and correlation.
ggstatsplot: Used for creating graphics with details from statistical tests included
heatmaply: Used to plot ‘heatmap’, a popular graphical method for visualizing high-dimensional data
parallelPlot: Used to create a parallel coordinates plot
Show code
packages = c('tidyverse','knitr', 'corrplot', 'ggstatsplot', 'rmarkdown', 'heatmaply', 'dendextend','parallelPlot')
for(p in packages){
if(!require(p, character.only = T)){
install.packages(p)
}
library(p, character.only = T)
}
3 Dataset
For this task, the Starbucks Drinks dataset is used.
3.1 Data Preparation
Import and combine data sets
The dataset was imported using the read_csv() function.
| Category | Name | Portion(fl oz) | Calories | Calories from fat | Caffeine(mg) | Size | Milk | Whipped Cream |
|---|---|---|---|---|---|---|---|---|
| tea | Iced Teavana® London Fog Tea Latte | 24 | 160 | 25 | 40–60 | Venti Iced | Almond | NA |
| tea | Iced Teavana® London Fog Tea Latte | 24 | 180 | 35 | 40–60 | Venti Iced | Coconut | NA |
| tea | Iced Teavana® London Fog Tea Latte | 24 | 180 | 0 | 40–60 | Venti Iced | Nonfat milk | NA |
| tea | Iced Teavana® London Fog Tea Latte | 24 | 230 | 50 | 40–60 | Venti Iced | Whole Milk | NA |
| tea | Iced Teavana® London Fog Tea Latte | 24 | 210 | 35 | 40 | Venti Iced | 2% Milk | NA |
| tea | Iced Teavana® London Fog Tea Latte | 24 | 210 | 25 | 40–60 | Venti Iced | Soy (United States) | NA |
Cleaning Caffeine(mg) Field
The field ‘Caffeing(mg)’ is classified as a string datatype as some cells contains a range of values. Rows containing ‘40+’ was first converted to ‘40’. Next, rows containing range i.e ‘25-40’, were identified and converted to the max. value i.e. ‘60’ using a for loop and ifelse() condition. Lastly, the column was converted to numeric datatype.
Show code
drinks["Caffeine(mg)"][drinks["Caffeine(mg)"] == '40+'] <- '40'
for (i in 1:nrow(drinks)) {
drinks[i, "Caffeine(mg)"] <- ifelse(grepl("–", drinks[i, "Caffeine(mg)"]),
substr(drinks[i, "Caffeine(mg)"],
nchar(drinks[i, "Caffeine(mg)"])-1,
nchar(drinks[i, "Caffeine(mg)"])),
drinks[i, "Caffeine(mg)"]
)
}
drinks["Caffeine(mg)"] <- as.numeric(unlist(drinks["Caffeine(mg)"]))
kable(tail(drinks[,c(1:2, 15)]))
| Category | Name | Caffeine(mg) |
|---|---|---|
| tea | Iced Teavana® London Fog Tea Latte | 60 |
| tea | Iced Teavana® London Fog Tea Latte | 60 |
| tea | Iced Teavana® London Fog Tea Latte | 60 |
| tea | Iced Teavana® London Fog Tea Latte | 60 |
| tea | Iced Teavana® London Fog Tea Latte | 40 |
| tea | Iced Teavana® London Fog Tea Latte | 60 |
3.2 Data Wrangling
Identifying Top 4 Largest Categories of Drinks
To identify the largest drink categories, the group_by() function was used to group the orders by category and summarise() was used to count (i.e. n()) the total number of drinks for each category. Then, arrange(desc) was used to sort the data and top_n() was used to select and identify the top 4 largest categories. filter() was used to display rows of drinks that are in the top 4 largest categories. They are espresso, frappuccino blended beverages, kids and others, and tea.
Show code
| Category | Name | Portion(fl oz) | Calories | Calories from fat | Caffeine(mg) | Size | Milk | Whipped Cream |
|---|---|---|---|---|---|---|---|---|
| espresso | Iced Starbucks® Blonde Caffè Latte | 12 | 50 | 30 | 85 | Tall | Almond | NA |
| espresso | Iced Starbucks® Blonde Caffè Latte | 12 | 70 | 35 | 85 | Tall | Coconut | NA |
| espresso | Iced Starbucks® Blonde Caffè Latte | 12 | 70 | 0 | 85 | Tall | Nonfat milk | NA |
| espresso | Iced Starbucks® Blonde Caffè Latte | 12 | 120 | 50 | 85 | Tall | Whole Milk | NA |
| espresso | Iced Starbucks® Blonde Caffè Latte | 12 | 100 | 35 | 85 | Tall | 2% Milk | NA |
| espresso | Iced Starbucks® Blonde Caffè Latte | 12 | 100 | 25 | 85 | Tall | Soy (United States) | NA |
Identifying Top 3 Largest Drink Names
To identify the largest drink names, the group_by() function was used to group the orders by name and summarise() was used to count (i.e. n()) the total number of drinks for each name. Then, arrange(desc) was used to sort the data and top_n() was used to select and identify the top 3 largest names. filter() was used to display rows of drinks that are in the top 3 largest names. They are iced coffee, hot chocolate and pumpkin spice crème.
Show code
| Category | Name | Portion(fl oz) | Calories | Calories from fat | Caffeine(mg) | Size | Milk | Whipped Cream |
|---|---|---|---|---|---|---|---|---|
| iced-coffee | Iced Coffee with Milk | 30 | 35 | 20 | 240 | Trenta Iced | Almond | Unsweetened |
| iced-coffee | Iced Coffee with Milk | 30 | 210 | 25 | 190 | Trenta Iced | Coconut | Sweetened |
| iced-coffee | Iced Coffee with Milk | 30 | 50 | 25 | 240 | Trenta Iced | Coconut | Unsweetened |
| iced-coffee | Iced Coffee with Milk | 30 | 210 | 0 | 190 | Trenta Iced | Nonfat milk | Sweetened |
| iced-coffee | Iced Coffee with Milk | 30 | 50 | 0 | 240 | Trenta Iced | Nonfat milk | Unsweetened |
| iced-coffee | Iced Coffee with Milk | 30 | 240 | 40 | 190 | Trenta Iced | Whole Milk | Sweetened |
Normalising Against Volume of Drink
As the nutritional value of the drinks contain different toppings i.e. milk and whipped cream, and are also based on different volumes and serving sizes, we will normalise the nutritional value by the volume of drink for each milk and whipped cream type.
The group_by() function was used to group the orders by name, milk and whipped cream and summarise() was calculate the nutritional value of the drink per unit volume. Next, the name, milk and whipped cream column was joined using paste() to create a single full name for the drink. The full name was then assigned to be the row name of the dataset using row.names()
Show code
topdrinks_norm <- topdrinks %>%
group_by(`Name`, `Milk`, `Whipped Cream`) %>%
summarise('Calories per oz'= mean(`Calories`/`Portion(fl oz)`),
'Calories from fat per oz'= mean(`Calories from fat`/`Portion(fl oz)`),
'Total Fat(g/oz)'= mean(`Total Fat(g)`/`Portion(fl oz)`),
'Saturated fat(g/oz)'= mean(`Saturated fat(g)`/`Portion(fl oz)`),
'Trans fat(g/oz)'= mean(`Trans fat(g)`/`Portion(fl oz)`),
'Cholesterol(mg/oz)'= mean(`Cholesterol(mg)`/`Portion(fl oz)`),
'Sodium(mg/oz)'= mean(`Sodium(mg)`/`Portion(fl oz)`),
'Total Carbohydrate(g/oz)'= mean(`Total Carbohydrate(g)`/`Portion(fl oz)`),
'Dietary Fiber(g/oz)'= mean(`Dietary Fiber(g)`/`Portion(fl oz)`),
'Sugars(g/oz)'= mean(`Sugars(g)`/`Portion(fl oz)`),
'Protein(g/oz)'= mean(`Protein(g)`/`Portion(fl oz)`),
'Caffeine(mg/oz)'= mean(`Caffeine(mg)`/`Portion(fl oz)`)) %>%
ungroup()
topdrinks_norm$Full_Name <- paste(topdrinks_norm$Name, topdrinks_norm$Milk, topdrinks_norm$`Whipped Cream`)
topdrinks_n <- topdrinks_norm %>%
select(c(4:15))
row.names(topdrinks_n) <- topdrinks_norm$Full_Name
kable(head(topdrinks_n[,c(1:7)]))
| Calories per oz | Calories from fat per oz | Total Fat(g/oz) | Saturated fat(g/oz) | Trans fat(g/oz) | Cholesterol(mg/oz) | Sodium(mg/oz) |
|---|---|---|---|---|---|---|
| 20.86667 | 5.000000 | 0.5641667 | 0.3700000 | 0.000 | 1.745833 | 10.083333 |
| 26.83333 | 10.075000 | 1.1066667 | 0.6783333 | 0.005 | 3.537500 | 10.641667 |
| 15.86667 | 4.358333 | 0.5000000 | 0.1250000 | 0.000 | 0.000000 | 9.441667 |
| 21.56667 | 9.183333 | 1.0075000 | 0.4283333 | 0.000 | 1.979167 | 10.083333 |
| 17.47500 | 5.641667 | 0.6383333 | 0.5641667 | 0.000 | 0.000000 | 8.883333 |
| 23.34167 | 10.175000 | 1.1458333 | 0.8791667 | 0.000 | 1.979167 | 9.441667 |
4 Visualisation
In this section, we will explore data visualisations for multidimensional data using:
- Corrgram
- Heatmap
- Parallel Coordinate Plot
4.1 Corrgram
A corrgram is a visual display technique that helps us to represent the pattern of relations among a set of variables in terms of their correlations.
4.1.1 Corrgram using pairs()
The figure below shows a conventional corrgram using the pairs() function.
Show code
panel.cor <- function(x, y, digits=2, prefix="", cex.cor, ...) {
usr <- par("usr")
on.exit(par(usr))
par(usr = c(0, 1, 0, 1))
r <- abs(cor(x, y, use="complete.obs"))
txt <- format(c(r, 0.123456789), digits=digits)[1]
txt <- paste(prefix, txt, sep="")
if(missing(cex.cor)) cex.cor <- 0.8/strwidth(txt)
text(0.5, 0.5, txt, cex = cex.cor * (1 + r) / 2)
}
pairs(drinks[,4:15],
upper.panel = panel.cor,
label.pos = 0.5,
line.main = 3,
cex.labels = 0.5,
font.labels = 0.5,
gap = 0.2)
4.1.2 Corrgram using ggstatplot()
The figure below shows a corrgram using the ggstatplot package. ggcorrplot and ggplot elements can be added into the the corrgram to customise the corrgram’s size and colour.
Show code
ggstatsplot::ggcorrmat(
data = drinks,
cor.vars = 4:15,
ggcorrplot.args = list(outline.color = "black",
hc.order = TRUE,
lab_col = "black",
lab_size = 2,
pch.col = "red",
pch.cex = 6),
title = "Correlogram for Starbucks Drink dataset",
subtitle = "One pair is not significant at p < 0.05",
ggplot.component = list(theme_void(base_size = 9),
theme(plot.title=element_text(size=12),
plot.subtitle=element_text(size=9),
legend.text = element_text(size=6),
axis.text.x = element_text(size = 6, angle = 45, hjust = 0.6),
axis.text.y = element_text(size = 6, hjust = 1)
))
)
4.1.3 Multiple Corrgrams using ggstatplot()
The figure below shows multiple corrgrams using the ggstatplot() function.
Show code
grouped_ggcorrmat(
data = topcat,
cor.vars = 4:15,
grouping.var = Category,
type = "p",
p.adjust.method = "holm",
plotgrid.args = list(ncol = 2),
ggcorrplot.args = list(outline.color = "black",
lab_col = "black",
lab_size = 1.5,
pch.col = "red",
pch.cex = 3),
annotation.args = list(tag_levels = "a",
title = "Correlogram for Top 4 Categories of Starbucks Drink dataset",
subtitle = "The categories are: Espresso, Frapuccino blended beverages, Kids Drinks and Tea"),
ggplot.component = list(theme_void(base_size = 7),
theme(plot.title = element_text(size=5),
plot.subtitle = element_text(size=3),
legend.text = element_text(size=5),
axis.text.x = element_text(size = 5, angle = 45, hjust = 0.6),
axis.text.y = element_text(size = 5, hjust = 1),
strip.text.x = element_text(size = 7),
legend.key.size = unit(3, 'mm')
))
)
4.1.4 Corrgram with significant level of 0.1 using corrplot()
The figure below shows a corrgram using the corrplot combined with the significant test of 0.1. The corrgram reveals that not all correlation pairs are statistically significant. For example the correlation between total carbohydrate and sugar is statistically significant at significant level of 0.1 but not the pair between total caffeine and trans fat.
Show code
drinks.cor <- cor(drinks[, 4:15])
drinks.sig = cor.mtest(drinks.cor, conf.level= .9)
corrplot.mixed(drinks.cor,
lower = "number",
upper = "square",
order="AOE",
tl.pos = "lt",
tl.col = "black",
tl.cex = .6,
tl.srt = 45,
pch.col = "grey70",
pch.cex = 1.5,
number.cex = .6,
cl.cex = .6,
lower.col = "black",
p.mat = drinks.sig$p,
sig.level = 0.1,
title = "Correlogram of Starbucks Drinks with significant level of 0.1",
mar=c(0,0,1,0)
)
4.1.5 Corrgram with hierarchical clustering
The dend_expend() and find_k() functions of dendextend package was used to determine the best clustering method and number of cluster.
Show code
drinks_matrix <- data.matrix(drinks.cor)
drinks_d <- dist(normalize(drinks_matrix[c(1:12)]), method = "euclidean")
drinks_clust <- hclust(drinks_d, method = "average")
num_k <- find_k(drinks_clust)
plot(num_k)
Next, the corrgram was plotted using corrplot() and hclust based on the results of hierarchical clustering.
Show code
corrplot(drinks.cor,
method = "ellipse",
order="hclust",
hclust.method = "ward.D",
addrect = 3,
tl.pos = "lt",
tl.col = "black",
tl.cex = .6,
tl.srt = 45,
number.cex = .6,
cl.cex = .6,
title = "Correlogram of Starbucks Drinks with 3 levels of hierarchical cluster",
mar=c(0,0,1,0))
4.1.6 Conclusion
In general, the corrgram for all starbucks drinks show that caffeine is mostly negatively correlated with the other nutritional factors except protein, whereas the rest are positively correlated. The diagram also shows that following pairs of nutritional factors of starbucks drinks are highly correlated (r > 0.90):
- Total Fat(g) – Calories from Fat (r = 1)
- Total Fat(g) – Saturated Fat(g) (r = 0.94)
- Calories from fat – Saturated Fat (g) (r = 0.94)
- Sugars (g) – Total Carbohydrate (g) (r = 0.99)
- Sugars (g) – Calories (r = 0.92)
- Total Carbohydrate (g) – Calories (r = 0.94)
The trans fat(g) and caffeine (mg) pair is not significant at p <0.05 and has a correlation parameter of only 0.01.
From the multiple corrgram, an interesting finding is that the caffeine for kids drinks and tea is positively correlated to the other factors.
The starbucks drinks nutrition factors can be separated into 3 clusters:
- Caffeine (mg)
- Trans fat(g), Cholesterol(mg), Saturated fat(g), Calories from fat, Total Fat(g)
- Sodium (mg), Calories, Total Carbohydrate(g), Sugars(g), Dietary Fiber(g), Protein(g)
The nutrition factors in each cluster are correlated with one another. Caffeine is standalone as it is not highly correlated with the others and generally has a negative correlation with the rest.
4.2 Heatmap
A heatmap is a data visualization technique that shows magnitude of a phenomenon as color in two dimensions. The variation in color may be by hue or intensity, giving obvious visual cues to the reader about how the phenomenon is clustered or varies over space.
The top drinks dataset where nutritional values have been normalised against the unit volume will be used for plotting the heat map.
First, the dend_expend() and find_k() functions of dendextend package was used to determine the best clustering method and number of cluster.
Show code
topdrinks_matrix <- data.matrix(topdrinks_n)
topdrinks_d <- dist(normalize(topdrinks_matrix[c(1:12)]), method = "euclidean")
topdrinks_clust <- hclust(topdrinks_d, method = "average")
topnum_k <- find_k(topdrinks_clust)
plot(topnum_k)
Next, heatmaply package was used to plot the heatmap for Ice Coffee, Hot Chocolate and Pumpkin Spice Crème for different combinations of milk and whipped cream.
Show code
4.2.1 Conclusion
The heatmap compares the nutritional value of hot chocolate, pumpkin spice crème and iced coffee, which are popular drinks in Starbucks. It shows that hot chocolate and pump spice crème are generally unhealthier, containing higher sodium, sugars, carbohydrates, and cholesterol levels than iced coffee. On the other hand, iced coffee contains higher caffeine levels than the hot chocolate and pump spice crème. The impact of milk, whipped cream, sweetener choices on the nutritional value of drinks were further analysed using hierarchical clustering. The drinks were separated into 4 clusters:
- Hot Chocolate and Pumpkin Spice Crème with whipped cream or Pumpkin Spiced Crème with Whole/2% milk
- Pumpkin Spice Crème without whipped cream with soy or nonfat milk
- Hot Chocolate and Pumpkin Spice Crème without whipped cream
- Iced Coffee with all milk and sweetener combinations
For Hot Chocolate and Pumpkin Spice Crème, the nutritional value was determined by whipped cream then milk type. In general, no whipped cream and plant-based milk milk choices are considered healthier with lower sodium, sugars, carbohydrates, and cholesterol levels. For Iced coffee, the nutritional value was determined by sweetener then milk type. Unsweetened iced coffee with plant-based milk is considered healthier with lower sodium, sugars, carbohydrates, and cholesterol levels.
4.3 Parallel Coordinate Plot
Parallel coordinates are a common way of visualizing and analyzing high-dimensional datasets. To show a set of points in an n-dimensional space, a backdrop is drawn consisting of n parallel lines, typically vertical and equally spaced. A point in n-dimensional space is represented as a polyline with vertices on the parallel axes; the position of the vertex on the i-th axis corresponds to the i-th coordinate of the point.
The parallel coordinate was plotted using the parallelPlot package.
Show code
4.3.1 Conclusion
The findings from the parallel coordinate plot are generally in line with the corrgram above. Drinks with high calories typically have high total carbohydrate, sugars, sodium and lower caffeine, vice versa. Some factors like trans fat, dietary fibre and cholesterol are generally not-well distributed with most drinks having a low nutritional value for those factors. They may not be a good indicator of the calorific content of the drinks.