ccf: Canonical correlation forest

In jandob/ccf: Canonical correlation forest

Description

This function computes a classifier based on a canonical correlation forest. It expects its input in matrix form or as formula notation.

Usage

canonical_correlation_forest(x, y = NULL, ntree = 200, verbose = FALSE,
  ...)

## Default S3 method:
canonical_correlation_forest(x, y = NULL, ntree = 200,
  verbose = FALSE, projectionBootstrap = FALSE, ...)

## S3 method for class 'formula'
canonical_correlation_forest(x, y = NULL, ntree = 200,
  verbose = FALSE, ...)

Arguments

x	Numeric matrix (n * p) with n observations of p variables
y	Numeric matrix with n observations of q variables
ntree	Number of trees the forest will be composed of
verbose	Optional argument to control if additional information are printed to the output. Default is FALSE.
...	Further arguments passed to or from other methods.
projectionBootstrap	Use projection bootstrapping. (default FALSE)

Value

returns an object of class "canonical_correlation_forest", where an object of this class is a list containing the following components:

• x,yThe original input data

• y_encodedThe encoded y variable in case of classification tasks.

• foresta vector of length ntree with objects of class canonical_correlation_tree.

References

Rainforth, T., and Wood, F. (2015): Canonical correlation forest, arXiv preprint, arXiv:1507.05444, https://arxiv.org/pdf/1507.05444.pdf.

Examples

data(spirals)

d_train <- spirals[1:1000, ]
d_test <- spirals[-(1:1000), ]

# compute classifier on training data
## variant 1: matrix input
m1 <- canonical_correlation_forest(d_train[, c("x", "y")], d_train$class, ntree = 20)
## variant 2: formula notation
m2 <- canonical_correlation_forest(class ~ ., d_train)

# compute predictive accuracy
get_missclassification_rate(m1, d_test)
get_missclassification_rate(m2, d_test)

jandob/ccf documentation built on May 18, 2019, 12:23 p.m.