bestm: Function called by makewpstRO to identify which packets are individually good for correlating with a response

Description

This function is used when you have a huge number of packets where you want to identify which ones are, individually, candidates for the good prediction of a response

Usage

bestm(w2mobj, y, percentage = 50)

Arguments

w2mobj

The w2m object that contains the packets you wish to preselect

The response time series

percentage

The percentage of the w2m packets that you wish to select

Value

A list of class w2m with the following components:
mA matrix containing the select packets (as columns), reordered so that the best packets come first
ixvecA vector which indexes the best packets into the original supplied matrix
pktixThe original wavelet packet indices corresponding to each packet
levelAs pktix but for the wavelet packet levels
nlevelsWTThe number of resolution levels in the original wavelet packet object
cvThe ordered correlations

Details

This function naively addresses a very common problem. The object w2mobj contains a huge number of variables which might shed some light on the response object y. The problem is that the dimensionality of w2mobj is larger than that of the length of the series y.

The solution here is to choose a large, but not huge, subset of the variables that might be potentially useful in correlating with y, discard the rest, and return the "best" or preselected variables. Then the dimensionality is reduced and more sophisticated methods can be used to perform better quality modelling of the response y on the packets in w2mobj.