Ensemble Sparse Partial Least Squares for Outlier Detection

Outlier detection with ensemble sparse partial least squares.

Usage

enspls.od(
  x,
  y,
  maxcomp = 5L,
  cvfolds = 5L,
  alpha = seq(0.2, 0.8, 0.2),
  reptimes = 500L,
  method = c("mc", "boot"),
  ratio = 0.8,
  parallel = 1L
)

Arguments

x

Predictor matrix.

y

Response vector.

maxcomp

Maximum number of components included within each model. If not specified, will use 5 by default.

cvfolds

Number of cross-validation folds used in each model for automatic parameter selection, default is 5.

alpha

Parameter (grid) controlling sparsity of the model. If not specified, default is seq(0.2, 0.8, 0.2).

reptimes

Number of models to build with Monte-Carlo resampling or bootstrapping.

method

Resampling method. "mc" (Monte-Carlo resampling) or "boot" (bootstrapping). Default is "mc".

ratio

Sampling ratio used when method = "mc".

parallel

Integer. Number of CPU cores to use. Default is 1 (not parallelized).

Value

A list containing four components:

• error.mean - error mean for all samples (absolute value)

• error.median - error median for all samples

• error.sd - error sd for all samples

• predict.error.matrix - the original prediction error matrix

Note

To maximize the probablity that each observation can be selected in the test set (thus the prediction uncertainty can be measured), please try setting a large reptimes.

See also

See enspls.fs for measuring feature importance with ensemble sparse partial least squares regressions. See enspls.fit for fitting ensemble sparse partial least squares regression models.

Author

Nan Xiao <https://nanx.me>

Examples

data("logd1k")
x <- logd1k$x
y <- logd1k$y

set.seed(42)
od <- enspls.od(
  x, y,
  reptimes = 5, maxcomp = 3,
  alpha = c(0.3, 0.6, 0.9)
)
plot(od, prob = 0.1)

plot(od, criterion = "sd", sdtimes = 1)