BKPC (version 1.0.1)

predict.bkpc: Predict Method for Bayesian Kernel Projection Classifier

Description

This function predicts values based upon a model trained by bkpc for new input data. BKPC employs a Gibbs sampler to sample from the posterior distributions of the parameters, so sample probability distributions of prediction can be obtained for for new data points.

Usage

# S3 method for bkpc
predict(object, newdata  = NULL, n.burnin = 0, ...)

Arguments

object

a bkpc object.

newdata

a matrix containing the new input data

n.burnin

number of burn-in iterations to discard.

…

Currently not used.

Value

A list with the following components:

class

estimated class for each observation in the input data.

map

maximum a posteriori probability estimate for belonging to each class for all observations in the input data.

p

a matrix of samples of estimated probabilities for belonging to each class for each observation in the input data.

Details

If newdata is omitted the predictions are based on the data used for the fit.

References

Domijan K. and Wilson S. P.: Bayesian kernel projections for classification of high dimensional data. Statistics and Computing, 2011, Volume 21, Issue 2, pp 203-216

See Also

bkpc

Examples

# NOT RUN {
set.seed(-88106935)

data(iris)
testset <- sample(1:150,30)

train <- as.matrix(iris[-testset,-5])
test <- as.matrix(iris[testset,-5])

wtr <- iris[-testset, 5]
wte <- iris[testset, 5]

result <- bkpc(train, y = wtr,  n.iter = 1000,  thin = 10, n.kpc = 2, 
intercept = FALSE, rotate = TRUE)

# predict
out <- predict(result, test, n.burnin = 20) 

# classification rate for the test set

sum(out$class == as.numeric(wte))/dim(test)[1]

table(out$class, as.numeric(wte))

# consider just misclassified observations:

missclassified <- out$class != as.numeric(wte)


tab <- cbind(out$map[missclassified, ], out$class[missclassified],  as.numeric(wte)[missclassified])
colnames(tab) = c("P(k = 1)", "P(k = 2)", "P(k = 3)", "predicted class", "true class")
tab


# consider, say, 28th observation in the test set:
# sample probability distributions of belonging to each of the three classes: 


ProbClass2samples <- out$p[28, ]
ProbClass3samples <- out$p[28 + dim(test)[1], ]
ProbClass1samples <- 1 - (ProbClass2samples + ProbClass3samples)
hist(ProbClass1samples)
hist(ProbClass2samples)
hist(ProbClass3samples)


# }