test_custom

#!/usr/bin/env Rscript

rm(list=ls())
## * Package setup
library(methods, quietly=TRUE)  ## hasArgs is not loaded by default by Rscript
library(devtools, quietly=TRUE)
pkg_path <-  "."
try(unload(pkg_path))
document(pkg_path)
load_all(pkg_path)
test(pkg_path)

## * Loading data set
data(mini_mnist)
feats <- mini_mnist$train$images
targets <- mini_mnist$train$labels

feats_oos <- mini_mnist$test$images
targets_oos <- mini_mnist$test$labels


## * Defining model
## ** initial weights
weights_init <- matrix(rnorm(NCOL(feats) * NCOL(targets)), NCOL(feats),
                       NCOL(targets))
## * Training model
## ** Setting training variables
max_iter <- Inf
decay <- 0
step_size <- 0.01
verbose <- FALSE
tol <- 1e-6

## ** Training models and outputing misclassification rates
for (backend in c("R", "C", "CUDA")) {
    model <- Classifier(weights_init=weights_init)
    model <- train(model, feats, targets, decay=decay, step_size=step_size,
                   max_iter=max_iter, verbose=verbose,
                   tol=tol, backend=backend)
    mis_is <- get_error(model, feats, targets)
    mis_oos <- get_error(model, feats_oos, targets_oos)
    print(paste(backend, "backend in sample misclassification rate"))
    print(mis_is)
    print(paste(backend, "backend out of sample misclassification rate"))
    print(mis_oos)
}