R | `neuralnet` Package
The neuralnet package for R offers a quick way to create simple neural nets for educational purposes:
library(neuralnet)training_data <- read.csv("https://raw.githubusercontent.com/jcha-ultra/data_toolkit/master/datasets/roots.csv", header = TRUE)model <- neuralnet(formula = root_k ~ k, data = training_data, hidden = 10)output <- cbind(training_data, nn_output = unlist(model$net.result))
However, this package is not suitable for production-level modeling work, for the following reasons:
it is not optimized for speed, as the package documentation itself states
it has limited options for configuring the neural net's hyperparameters and architecture
it does not provide the option to expose the internal workings of the neural net and make fine adjustments to it
A more full-fledged framework for creating neural nets is Keras, which is built on top of Google's Tensorflow and has an R interface [2].
Footnotes/Resources:
[3] A longer demo of fitting a model: https://www.r-bloggers.com/2015/09/fitting-a-neural-network-in-r-neuralnet-package
[4] Notebook file (by this card's author) showing exploration of neural net fitting using this package: https://github.com/jcha-ultra/data_toolkit/blob/master/learning_projects/titanic_nn/titanic_neural_net.Rmd