predict.gspca: Predict response with a generalized supervised PCA model

In andland/genSupPCA: Supervised Dimensionality Reduction for Exponential Family Data

Description

Predict response with a generalized supervised PCA model

Usage

## S3 method for class 'gspca'
predict(object, newdata, type = c("link", "response", "PCs"),
  ...)

Arguments

object	generalized supervised PCA object
newdata	matrix of the same exponential family as covariates in object. If missing, will use the data that object was fit on
type	the type of fitting required. type = "link" gives response variable on the natural parameter scale, type = "response" gives response variable on the response scale, and type = "response" gives matrix of principal components of x
...	Additional arguments

Examples

# construct a low rank matrices in the natural parameter space
rows = 100
cols = 10
set.seed(1)
loadings = rnorm(cols)
mat_np = outer(rnorm(rows), rnorm(cols))
mat_np_new = outer(rnorm(rows), loadings)

# generate a count matrices and binary responses
mat = matrix(rpois(rows * cols, c(exp(mat_np))), rows, cols)
mat_new = matrix(rpois(rows * cols, c(exp(mat_np_new))), rows, cols)

response = rbinom(rows, 1, rowSums(mat) / max(rowSums(mat)))
response_new = rbinom(rows, 1, rowSums(mat_new) / max(rowSums(mat_new)))

mod = genSupPCA(mat, response, k = 1, family_x = "poisson", family_y = "binomial",
                quiet = FALSE, max_iters_per = 1, discrete_deriv = FALSE)

plot(predict(mod, mat, type = "response"), response_new)

andland/genSupPCA documentation built on May 30, 2019, 11:43 a.m.