rf_parcoor: Parallel coordinates plots of the paths through a random...
In brfitzpatrick/forestviews: Visualise the paths through random forests

Description Usage Arguments See Also Examples

rf_parcoor produces parallel coordinates plots of the paths through a random forest. These visualisations are introduced and explained in our manuscript, a preprint of which is available from arXiv at https://arxiv.org/abs/1706.08702.

1
2
3

rf_parcoor(all.paths.out, plot = TRUE, all.nodes = TRUE, plot.title = "",
  plot.x.nodes = 7, colour.scale.log = TRUE, colour.scale.lim = 9,
  grey.scale = TRUE)

`all.paths.out`	the result of supplying a random forest to `rf_pathfiner`.
`plot`	if `TRUE` a `ggplot2` plot object is returned that is the parallel coordinates plot. If `FALSE` the dataframe is returned from which a parallel coordinates plot can be produced with `ggplot2` (see examples below).
`all.nodes`	If `TRUE` the parallel coordinates plot depicts all nodes on all paths through the random forest. If `FALSE` the parallel coordinates plot depicts the first `plot.x.nodes` nodes along all paths through the random forest.
`plot.title`	a character string specifying the title to display on the plot.
`plot.x.nodes`	the number of nodes along all paths to represent with parallel vertical axes
`colour.scale.log`	If `TRUE` the colour scale for the path selection frequencies uses a logarithmic scale. If `FALSE` the colour bar does not use a logarithmic scale.
`colour.scale.lim`	The upper bound of the colour scale (if `colour.scale.log = TRUE` supply the logarithm of the upper bound desired to this argument).
`grey.scale`	If `TRUE` a grey scale is used in place of a colour scale to represent the path selection frequencies. If `FALSE` a colour scale is used to represent the path selection frequencies.

rf_pathfinder

# example 1:
library(mlbench)
data(Satellite)
library(randomForest)
rf.1 <- randomForest(classes ~ ., data = Satellite, mtry = 8, keep.forest = TRUE, ntree = 25, importance = TRUE)
rf.1.all.paths <- rf_pathfinder(rf = rf.1)
rf.pc <- rf_parcoor(all.paths.out = rf.1.all.paths, plot = TRUE, all.nodes = TRUE, plot.title = '', grey.scale = FALSE)
rf.pc
# custom ordering of covariates on the vertical axis:
rf.pc + ggplot2::ylim(paste(c('Terminus', paste('x',1:36,sep = '.'))))

# grey scale version:
rf.pc.grey <- rf_parcoor(all.paths.out = rf.1.all.paths, plot = TRUE, all.nodes = TRUE, plot.title = '', grey.scale = TRUE)
rf.pc.grey

# interactive version with plotly (should work once geom_GeomCurve() is implemented in plotly)
# library(plotly)
# ggplotly(p = p1)

# example 2:
data(iris)
library(randomForest)
rf.2 <- randomForest(Species ~ ., data = iris, mtry = 2, keep.forest = TRUE, ntree = 100, importance = TRUE)
rf.2.all.paths <- rf_pathfinder(rf = rf.2)
rf.pc.2 <- rf_parcoor(all.paths.out = rf.2.all.paths, plot = TRUE, all.nodes = TRUE, plot.title = '', grey.scale = TRUE)
rf.pc.2