gpsrc: Gaussian process for survival, regression and classification

In linxihui/sml: Machine Learning for Survival Data

Description

It works for classification, regression, cox regression and poission regression The first/main purpose is to implement Gaussian process into Cox's model The glmnet is used to solve a general Ridge-regularized regression solver. This may be changed in the future.

Usage

gpsrc(x, ...)

## S3 method for class 'formula'
gpsrc(formula, data = environment(formula), ...)

## Default S3 method:
gpsrc(x, y, scaled = TRUE, kernel = "rbfdot",
  kpar = "automatic", var = 1, family = switch(class(y), Surv = "cox",
  factor = ifelse(nlevels(y) > 2, "multinomial", "binomial"), integer =
  "poisson", "gaussian"), ...)

## S3 method for class 'gpsrc'
predict(object, newdata, type = c("response", "probabilities",
  "link", "risk"))

Arguments

x	Design matrix (NO intercept)
...	Further argument passed to internal functions
formula	Model formula
data	Data
y	Reponse vector of type double, integer, factor or Surv
scaled	Logical value indicating if to standardize x, y
kernel	String or 'kernel' object (see kernlab::gausspr)
kpar	A list of Kernel parameters or 'automatic' if a radial kernel specified
var	Variance of response (from the Gaussian process). Only for regression
family	Options are 'gaussian', 'binomial', 'multinomial', 'poisson', 'cox'
object	'gpsrc' object from gpsrc
newdata	Design matrix of test set
type	Type of output

Value

A 'gprsc' object

Examples

## classification
library(pROC);
library(mlbench);
data(Sonar);
i.tr <- sample(nrow(Sonar), size = 0.6*nrow(Sonar));

gp <- gpsrc(Class ~., data = Sonar[i.tr,], kpar = list(sigma = 0.01))
gp.pred <- predict(gp, Sonar[-i.tr, ], type = 'prob')
cat('AUC:', roc(Sonar$Class[-i.tr], gp.pred[, 2])$auc, '\n');

## regression

data(BostonHousing);
BH <- BostonHousing;
i.tr <- sample(nrow(BH), 200);

gp2 <- gpsrc(medv ~., data = BH[i.tr, ], kpar = list(sigma = 0.1));
gp2.pred <- predict(gp2, BH[-i.tr, ])
cat('RMSE:', sqrt(mean((BH$medv[-i.tr] - gp2.pred)^2)), '\n');

## survival

library(survival);
data(pbc, package = 'randomForestSRC');
pbc <- na.omit(pbc);
i.tr <- sample(nrow(pbc), 100);

gp <- gpsrc(Surv(days, status) ~., data = pbc[i.tr, ],
	kernel = 'laplacedot', kpar = list(sigma = 0.1));
gp.pred <- predict(gp, pbc[-i.tr, ])
cat('C-index:', survConcordance(Surv(days, status) ~ gp.pred, data = pbc[-i.tr, ])$concordance, '\n');

linxihui/sml documentation built on May 21, 2019, 6:39 a.m.