In RevoEnhancements/RevoEnhancements: Selected enhancements to Revolution Analytics RevoScaleR

Visualizing Decision Trees and Forest objects with Revolution R and the RevoEnhancements package

Our package RevoEnhancements on github now contains an enhancement to RevoScaler that enables plotting of rxDtree and rxDForest objects.

First, load the required packages:

library(RevoScaleR)
library(RevoEnhancements)
library(ggplot2)
rxOptions(
  computeContext=RxLocalParallel(),
  reportProgress=0
)

Plotting decision trees:

The Revolution function to create a decision tree is rxDTree. We use rxDTree to fit a partition tree of the diamonds data, restricting the maximum tree depth to 3, simply to make the plot stand out better.

### Create and plot an rxDTree object

frm <- formulaExpand(price ~ ., data=diamonds)
fit <- rxDTree(frm, diamonds, maxDepth=3)

Since the package RevoEnhancements exports a method for plot.rxDTree, we simply have to call plot(fit) to visualise the tree:

par(mar=c(0.2, 0.2, 1.5, 0.2))
par(mai=rep(0.1, 4))
plot(fit)

The function plot.rxDTree takes additional arguments to control the plot and text characteristics:

• plotArgs should be a list and gets passed to plot.rpart. Use this to control plot axes, titles, etc.
• textArg is also a list and gets passed to text.rpart. Use this to control the text on the plot, e.g. colour (col) and character size (cex).

For example, to set the text colour to blue and the character expansion to 0.7, use the textArgs argument:

plot(fit, 
     textArgs=list(col="blue", cex=0.7))

To also set the plot title, use the plotArgs argument:

plot(fit, 
     textArgs=list(col="blue", cex=0.7), 
     plotArgs=list(main="Decision tree"))

Plotting Decision Forests:

For comparison we fit a decision forest on the same diamonds data. This creates a rxDForest object. This object is a list and the individual trees are contained in the forest element.

### Create and plot an rxDForest object

fit <- rxDForest(frm, diamonds, maxDepth=3)
forest <- fit$forest

Now we can plot the individual trees. Since each individual tree of an rxDForest object is itself and rxDTree object, the plotting is identical to what we illustrated earlier:

par(mar=c(0.2, 0.2, 1.5, 0.2))
par(mai=rep(0.1, 4))
plot(forest[[1]], textArgs=list(col="blue", cex=0.7))

Again, you can take control of the plot with the plotArgs and textArgs arguments:

plot(forest[[1]], 
     textArgs=list(col="blue", cex=0.7), 
     plotArgs=list(main="Decision forest: Tree 1"))
plot(forest[[2]], 
     textArgs=list(col="blue", cex=0.7), 
     plotArgs=list(main="Decision forest: Tree 2"))

RevoEnhancements/RevoEnhancements documentation built on May 9, 2019, 9:46 a.m.