predict.binaryGP: Predictions of Binary Gaussian Process

In binaryGP: Fit and Predict a Gaussian Process Model with (Time-Series) Binary Response

Description

The function computes the predicted response and its variance as well as its confidence interval.

Usage

## S3 method for class 'binaryGP'
predict(object, xnew, conf.level = 0.95,
  sim.number = 101, ...)

Arguments

object	a class binaryGP object estimated by binaryGP_fit.
xnew	a testing matrix with dimension n_new by d in which each row corresponds to a predictive location.
conf.level	a value from 0 to 1 specifying the level of confidence interval. The default is 0.95.
sim.number	a positive integer specifying the simulation number for Monte-Carlo method. The default is 101.
...	for compatibility with generic method predict.

Value

mean	a matrix with dimension n_new by T displaying predicted responses at locations xnew.
var	a matrix with dimension n_new by T displaying predictive variances at locations xnew.
upper.bound	a matrix with dimension n_new by T displaying upper bounds with conf.level confidence level.
lower.bound	a matrix with dimension n_new by T displaying lower bounds with conf.level confidence level.
y_pred	a matrix with dimension n_new by T displaying predicted binary responses at locations xnew.

Author(s)

Chih-Li Sung <[e-mail]>

See Also

binaryGP_fit for estimation of the binary Gaussian process.

Examples

library(binaryGP)

#####      Testing function: cos(x1 + x2) * exp(x1*x2) with TT sequences      #####
#####   Thanks to Sonja Surjanovic and Derek Bingham, Simon Fraser University #####
test_function <- function(X, TT)
{
  x1 <- X[,1]
  x2 <- X[,2]

  eta_1 <- cos(x1 + x2) * exp(x1*x2)

  p_1 <- exp(eta_1)/(1+exp(eta_1))
  y_1 <- rep(NA, length(p_1))
  for(i in 1:length(p_1)) y_1[i] <- rbinom(1,1,p_1[i])
  Y <- y_1
  P <- p_1
  if(TT > 1){
    for(tt in 2:TT){
      eta_2 <- 0.3 * y_1 + eta_1
      p_2 <- exp(eta_2)/(1+exp(eta_2))
      y_2 <- rep(NA, length(p_2))
      for(i in 1:length(p_2)) y_2[i] <- rbinom(1,1,p_2[i])
      Y <- cbind(Y, y_2)
      P <- cbind(P, p_2)
      y_1 <- y_2
    }
  }

  return(list(Y = Y, P = P))
}

set.seed(1)
n <- 30
n.test <- 10
d <- 2
X <- matrix(runif(d * n), ncol = d)
X.test <- matrix(runif(d * n.test), ncol = d)

##### without time-series #####
Y <- test_function(X, 1)$Y  ## Y is a vector
test.out <- test_function(X.test, 1)
Y.test <- test.out$Y
P.true <- test.out$P

# fitting
binaryGP.model <- binaryGP_fit(X = X, Y = Y)

# prediction
binaryGP.prediction <- predict(binaryGP.model, xnew = X.test)
print(binaryGP.prediction$mean)
print(binaryGP.prediction$var)
print(binaryGP.prediction$upper.bound)
print(binaryGP.prediction$lower.bound)

##### with time-series #####
Y <- test_function(X, 10)$Y  ## Y is a matrix with 10 columns
test.out <- test_function(X.test, 10)
Y.test <- test.out$Y
P.true <- test.out$P

# fitting
binaryGP.model <- binaryGP_fit(X = X, Y = Y, R = 1)

# prediction
binaryGP.prediction <- predict(binaryGP.model, xnew = X.test)
print(binaryGP.prediction$mean)
print(binaryGP.prediction$var)
print(binaryGP.prediction$upper.bound)
print(binaryGP.prediction$lower.bound)

binaryGP documentation built on May 1, 2019, 8:03 p.m.