Skip to main content

Community Contributions for EDAV Fall 2022 Mon/Wed

24 Tutorials for ggfortify and autoplotly

Yujia Chen

24.0.1 Motivation

The first time I used ggfortify package was to draw time series graph similar to ggplot2 style. ggfortify package has a very easy-to-use and uniform programming interface that enables users to use one line of code to visualize statistical results using ggplot2 as building blocks. The ggfortify package extends ggplot2 for plotting some popular R package using a standardized approach, included in the function autoplot().

The autoplotly package provides functionalities to automatically generate interactive visualizations for many popular statistical results supported by ggfortify package with plotly and ggplot2 style.

For example, we can hover the mouse over to each point in the plot to see more details, such as principal components information and the species this particular data point belongs to. We can also use the interactive selector to drag and select an area to zoom in, and zoom out by double clicking anywhere in the plot.

ggfortify and autoplotly can be widely used in statistical analysis fields and they can be seen as powerful additions to the ggplot2 package.

24.0.2 Examples

24.0.2.1 Linear Models

The autoplotly() function is able able to interpret lm fitted model objects and allows the user to select the subset of desired plots through the which parameter. The which parameter allows users to specify which of the subplots to display, for example:

autoplotly(
  lm(Petal.Width ~ Petal.Length, data = iris),
  which = c(4, 6), colour = "dodgerblue3",
  smooth.colour = "black", smooth.linetype = "dashed",
  ad.colour = "blue", label.size = 3, label.n = 5,
  label.colour = "blue")

24.0.2.2 Principal Component Analysis

The autoplotly() function works for the two essential classes of objects for principal component analysis (PCA) obtained from stats package: stats::prcomp and stats::princomp, for example:

pca <- autoplotly(prcomp(iris[c(1, 2, 3, 4)]), data = iris,
  colour = 'Species', frame = TRUE)
pca

24.0.2.3 Clustering

The autoplotly package also supports cluster::clara, cluster::fanny, cluster::pam as well as cluster::silhouette classes. It automatically infers the object type and generate interactive plots of the results from those packages with a single function call. Visualization of convex for each cluster in the iris data set:

autoplotly(fanny(iris[-5], 3), frame = TRUE)

By specifying frame, we are able to draw boundaries of different shapes. The different frame types can be found in ggplot2::stat_ellipse’s type keyword via frame.type option. Visualization of probability ellipse for iris data set:

autoplotly(pam(iris[-5], 3), frame = TRUE, frame.type = 'norm')

24.0.2.4 Forecasting

Forecasting packages such as forecast, changepoint, strucchange, and KFAS, are popular choices for statisticians and researchers. Interactive visualizations of predictions and statistical results from those packages can be generated automatically using the functions provided by autoplotly with the help of ggfortify.

The changepoint package provides a simple approach for identifying shifts in mean and/or variance in a time series. ggfortify supports cpt object in changepoint package.

Visualization of the change points with optimal positioning for the AirPassengers data set found in the changepoint package using the cpt.meanvar function:

autoplotly(cpt.meanvar(AirPassengers))

Visualization of the original and smoothed line in KFAS package:

model <- SSModel(
  Nile ~ SSMtrend(degree=1, Q=matrix(NA)), H=matrix(NA)
)

fit <- fitSSM(model=model, inits=c(log(var(Nile)),log(var(Nile))), method="BFGS")
smoothed <- KFS(fit$model)
autoplotly(smoothed)

Visualization of filtered result and smoothed line:

trend <- signal(smoothed, states="trend")
filtered <- KFS(fit$model, filtering="mean", smoothing='none')
p <- autoplot(filtered)
autoplotly(trend, ts.colour = 'blue', p = p)

Visualization of optimal break points where possible structural changes happen in the regression models built by strucchange::breakpoints:

autoplotly(breakpoints(Nile ~ 1), ts.colour = "blue", ts.linetype = "dashed",
           cpt.colour = "dodgerblue3", cpt.linetype = "solid")

24.0.2.5 Splines

The autoplotly can also automatically generate interactive plots for results produced by splines.

B-spline basis points visualization for natural cubic spline with boundary knots produced from splines::ns:

autoplotly(ns(ggplot2::diamonds$price, df = 6))

24.0.3 Extensibility and Composability

The plots generated using autoplotly() can be easily extended by applying additional ggplot2 elements or components. For example, we can add title and axis labels to the originally generated plot using ggplot2::ggtitle and ggplot2::labs:

autoplotly(
  prcomp(iris[c(1, 2, 3, 4)]), data = iris, colour = 'Species', frame = TRUE) +
  ggplot2::ggtitle("Principal Components Analysis") +
  ggplot2::labs(y = "Second Principal Component", x = "First Principal Component")

Similarly, we can add additional interactive components using plotly. The following example adds a custom plotly annotation element placed to the center of the plot with an arrow:

p <- autoplotly(prcomp(iris[c(1, 2, 3, 4)]), data = iris,
  colour = 'Species', frame = TRUE)

p %>% plotly::layout(annotations = list(
  text = "Example Content",
  font = list(
    family = "Courier New, monospace",
    size = 18,
    color = "black"),
  x = 0,
  y = 0,
  showarrow = TRUE))

We can also stack multiple plots generated from autoplotly() together in a single view using subplot(), two interactive splines visualizations with different degree of freedom are stacked into one single view in the following example:

subplot(
  autoplotly(ns(ggplot2::diamonds$price, df = 6)),
  autoplotly(ns(ggplot2::diamonds$price, df = 3)), nrows = 2, margin = 0.03)

This tutorial provides a brief introduction for interactive data visualization tools. As we can see, ggfortify and autoplotly generate interactive visualizations for many popular statistical results. I learned how to build beautiful visualizations of statistical results with concise code. In the future, I will try to plot more statistical models with ggfortify and autoplotly.