matr

A Simple R Package for Estimating Neural Network Models using a C++ Backend

Introduction

This R package provides a simple API for estimating neural network models in R using a C++ backend via Rcpp. The package was originally designed as my project for ISYE 6740.

Getting Started

To get started simply install the package from GitHub:

devtools::install_github("https://github.com/walkerjameschris/matr")

To train a model, load the package and data. The fit_network() function accepts the training data and labels (one hot encoded) as matrices. You can tune the number of hidden layer neurons (neurons) in addition to the learning rate (learn_rate), the max number of iterations (epoch), and a random seed.

data <- matr:::mnist

network <-
  matr::fit_network(
    X = data$X,
    Y = data$Y,
    neurons = 5,
    epoch = 1000,
    learn_rate = 0.0001
  )

predict(network)