predictML.predkmeans {predkmeans}

Prediction of Cluster Membership

Description

Predicts cluster membership using either multinomial logistic regression or SVMs.

Usage

## S3 method for class 'predkmeans'
predictML(
  object = NULL,
  centers = object$centers,
  K = nrow(centers),
  R,
  Rstar,
  Xstar = NULL,
  tr.assign = object$cluster,
  muStart = "random",
  maxitMlogit = 500,
  verbose = 1,
  nMlogitStarts = 1,
  mlogit.control = list(suppressFittedWarning = TRUE),
  ...
)

## S3 method for class 'predkmeans'
predictSVM(
  object = NULL,
  centers = object$centers,
  R,
  Rstar,
  K = nrow(centers),
  Xstar = NULL,
  tr.assign = object$cluster,
  svm.control = list(gamma = c(1/(2:1), 2), cost = seq(20, 100, by = 20)),
  ...
)

## S3 method for class 'predkmeans'
predictMixExp(object, R, Rstar = NULL, ...)

Arguments

object

A predkmeans object, from which the cluster centers will be extracted.

centers

Matrix of cluster centers, assumed to be K-by-p

K

Number of clusters

R

matrix of covariates for observations to be predicted at.

Rstar

matrix of covariates at training locations

Xstar

matrix of observation at training locations. Either this or tr.assign is required.

tr.assign

vector of cluster assignments at training locations. By default, extracted from object.

muStart

starting value for cluster centers in mlogit optimization (IDEA: change to pull from predkmeans object?). If not provided, starting values are selected randomly.

maxitMlogit

Maximum number of iterations for mlogit in prediction

verbose

integer indicating amount of output to be displayed

nMlogitStarts

number of mlogit starts to use in estimation of parameters

mlogit.control

list of control parameters to be passes to mlogit

...

Unused additional arguments

svm.control

list of options for best.svm

Details

Function for predicting cluster membership in clusters identified by k-means or predictive k-means using multinomial logistic regression or support vector machines (SVMs). For multinomial logitic regression, parameter estimation is handled by mlogit. The SVMs are fit using best.svm from e1071 package.

Because this prediction includes return information about cluster assignment and prediction model parameters, this method is deliberately distinct from the generic predict functions.

The predictMixExp funciton provides predictions from the 'working' cluster assignments created as part of the mixture of experts algorithm from predkmeans.

Value

A list containing some or all of the following elements:

tr.assign

Cluster assignments at training locations

mlfit

A subset of the mlogit object returned by the function of that name

beta

Estimated model parameters

test.pred

Predicted cluster assignments at test locations

Author(s)

Joshua Keller

See Also

mlogit, predkmeans, predictionMetrics

Other methods for predkmeans objects: relevel.predkmeans()

Examples

n <- 200
r1 <- rnorm(n)
r2 <- rnorm(n)
u1 <- rbinom(n, size=1,prob=0)
cluster <- ifelse(r1<0, ifelse(u1, "A", "B"), ifelse(r2<0, "C", "D"))
mu1 <- c(A=2, B=2, C=-2, D=-2)
mu2 <- c(A=1, B=-1, C=-1, D=-1)
x1 <- rnorm(n, mu1[cluster], 4)
x2 <- rnorm(n, mu2[cluster], 4)
R <- cbind(1, r1, r2)
X <- cbind(x1, x2)
pkm <- predkmeans(X=cbind(x1, x2), R=R, K=4)
n_pred <- 50
Rnew <- cbind(1, r1=rnorm(n_pred), r2=rnorm(n_pred))
pkmPred <- predictML(pkm, R=Rnew, Rstar=R)
pkmPred$test.pred