cwrEm: Function to estimate Cluster Weighted Regression (CWR) models

In cwm: Cluster Weighted Models by EM algorithm

Description

This function estimates CWR models via EM algoritms. An object of class cwrObj is returned containing posterior probabilities and group parameters.

Usage

cwrEm(X, Y, nc, max_iter = 1000, thresh = 0.01, cov_typeX = "full", 
cov_typeY = "full", clamp_weights = FALSE, create_init_params = TRUE, 
cwrStart = NULL, cov_priorX = NULL, cov_priorY = NULL, verbose = TRUE, 
regress = TRUE, clamp_covX = FALSE, clamp_covY = FALSE)

Arguments

X	X data matrix
Y	Y data matrix
nc	Number of clusters
max_iter	Max iterations. Default 1000
thresh	threshold to assess numerical convergence. Default 0.01
cov_typeX	Type of covariance of groups in X space. May be: "full" (default), "spherical", "diagonal"
cov_typeY	Type of covariance of groups in Y space. May be: "full" (default), "spherical", "diagonal"
clamp_weights	Fixed weights
create_init_params	Creates initial parameters
cwrStart	cwrObj to initialize. If autostart -> NULL
cov_priorX	Prior X covariance if not autostart. See cov_typeX
cov_priorY	Prior Y covariance if not autostart. See cov_typeY
verbose	Prints details of estimation process
regress	Regression model. Default TRUE
clamp_covX	Fixed covX matrix.
clamp_covY	Fixed covY matrix.

Details

This is the main function to estimate CWR models

Value

A CWR object with the following component:

muX	Means matrix of X component
muY	Means matrix of X component
aic	AIC of model
X	X matrix
Y	Y matrix
SigmaY	Array containing Y Variances
SigmaX	Array containing X Variances
weightsY	Matrix containing posterior probabilities

Warning

Estimation can be slow. Convergence is not guaranteeted.

Note

This is the main function. X and Y may be vectors or matrices. cwrObj objects containing parameters and posterior probabilities are returned.

Author(s)

Giorgio Spedicato

References

Murphy

See Also

stepCwr

Examples

##using Geyser dataset from package MASS
library(MASS)
data(geyser)
x=geyser[,1]
y=geyser[,2]                            
cwrEmExample=cwrEm(x,y,nc=2)
print(cwrEmExample) 

Example output

Loading required package: MASS
iteration  1  logLik  -3201.346 
iteration  2  logLik  -1419.224 
iteration  3  logLik  -1404.442 
iteration  4  logLik  -1401.795 
cwrObj :  cwrEmExample 
------------- 
 
Number of groups:  2 
Relative weights (priors):  0.6603206 0.3396794 
------------- 
 
logLik :  -1401.795 
AIC :  2827.59 
BIC :  2871.995 
------------- 
 
computation time :  0.653 

cwm documentation built on May 1, 2019, 9:19 p.m.