ddgraph (version 1.16.0)

ncpcResampling: NCPC Robustness from resampling

Description

Estimate the NCPC robustness using either jackknife or bootstrap resampling.

Usage

ncpcResampling(obj, method="bootstrap", method.param, verbose=TRUE, ...)

Arguments

obj

the DDDataSet object

method

the method to use to estimate how robust is the feature selection (valid values: "jackknife", or "bootstrap").

method.param

the parameter to method, either number of data points to remove for "jackknife" (default: 1) or number of boostrap runs for "bootstrap" (default: 100).

verbose

if to print out the progress

...

other parameters to pass to ncpc()

Value

Details

Estimate the robustness of NCPC predictions (i.e. variable types: direct, joint, indirect, no dependence) using resampling. Two type of resampling are available: bootstrap (where the whole dataset is resampled with replacement), and jackknifing (where 1 or more observation are removed at each resampling step).

NCPC is run for the resampled datasets and statistics is produced about how many times is each variable assigned one of the four types (direct, joint, indirect, no dependence). The final call for each variable is then made according to the following algorithm (#direct is number of times variable is called direct):

1. if #no dependence > #direct+joint+indirect => "no dependence"
2. else if #indirect > #direct+joint => "indirect"
3. else if #joint > #direct => "joint"
4. else "direct"