Dark Mode

NNetwork

A simple neural network in R as an R6 class object

Background

After learning R from a few R tutorials, I decided it was time to learn what machine learning and data science truly was. I had been putting off looking those up because I didn't feel like I was ready. I watched a series on neural networks by 3Blue1Brown on YouTube linked below.

But what is a neural network? | Chapter 1, Deep learning

"Oh, I can do that." I immediately thought to myself.

This video series applied old and familiar concepts of linear algebra and multivariable calculus that I had learned in college. Knowing that there were applications of this with data and programming inspired me to try to write some libraries from scratch.

Trained Neutal Network — My first trained neural network!

The Project

I chose R to do this rather than Python because I wanted to build experience with R. This project would be a good demonstration of mixing higher mathmatics with programming, which is what R was built to do.

I did not follow any programming tutorials when developing this. My primary intention was to familiarize myself with the math behind these concepts. I wrote out as much of the program that made sense to me and then referenced YouTube for more detailed topics as I came to them. I primariliy referenced a series by deeplizard on Neural Networks linked below.

Deep Learning Fundamentals - Intro to Neural Networks

This series was helpful and taught me about weights and bias initialization, the learning rate, and walked the tedious math sequences in a way that I could follow with my code.

One idea I came up with myself was simple testing process to verify the project worked. I decided to test the network by training it to read binary. This way I would not have to find or build a database of training data, nor label the data.

Next Steps

This project was written in December of 2022 and added to GitHub in February of 2023. The next seps I would like to do would be to formalize the testing process into a Unit Test with the testthat library in R. After that, I would like to format the library into a package that could be installed consistently into other machines. Not necesarrily through CRAN, but through GitHub.