How do I train a neural network using neural-net-ruby?

Use the provided example scripts like iris.rb for classification or mpg.rb for regression. Run the Ruby file with your data, and it will train the network using the Rprop algorithm as implemented in the code, with parameters adjustable in the script.

Is neural-net-ruby better than ruby-fann for machine learning?

Neural-net-ruby is better for learning due to its simplicity and pure Ruby code, while ruby-fann is superior for production use because it's faster and more optimized, being a wrapper for the C-based FANN library, as stated in the README.

What datasets can I use with neural-net-ruby?

The project includes built-in datasets for iris classification and car MPG regression in the examples. For OCR, you need to download the MNIST dataset separately, following instructions in mnist.rb, but custom datasets can be adapted by modifying the code.

How does the Rprop algorithm work in this implementation?

Rprop adjusts weights based only on the sign of the gradient, making training faster than standard backpropagation. This implementation follows the algorithm from linked sources, with code focused on simplicity for educational clarity.

Can neural-net-ruby handle deep learning projects?

No, it's designed for shallow feedforward neural networks and does not support deep architectures or advanced layers like CNNs. It's best for understanding basic concepts, not complex deep learning tasks.

What are the performance benchmarks for neural-net-ruby?

The README notes it trains faster than ai4r's backpropagation but is significantly slower than ruby-fann. Performance degrades with larger datasets, so it's not suited for benchmarks requiring high efficiency or speed.

Open-Awesome

neural-net-ruby

Ruby

A lightweight feedforward neural network with resilient backpropagation (Rprop), implemented in pure Ruby with no external dependencies.

GitHub

127 stars16 forks0 contributors

What is neural-net-ruby?

Neural Net (in Ruby) is a lightweight, educational implementation of a feedforward neural network using the resilient backpropagation (Rprop) training algorithm. It provides a simple, dependency-free way to understand neural network fundamentals in Ruby, with examples for classification, regression, and OCR tasks. The project is designed for learning and experimentation rather than production use.

Target Audience

Ruby developers and students interested in learning how neural networks work from scratch, or those seeking a minimal, readable implementation for educational purposes.

Value Proposition

Developers choose this for its simplicity, pure Ruby implementation with no dependencies, and clear examples that demystify neural network concepts. It offers a faster training alternative to basic backpropagation while remaining accessible and easy to modify.

Overview

A neural network, written in Ruby

Use Cases

Best For

Learning neural network fundamentals in Ruby
Experimenting with Rprop training algorithms
Educational projects on classification and regression
Understanding OCR with handwritten digits
Comparing neural network implementations
Building simple proof-of-concept models without external libraries

Not Ideal For

Production machine learning applications requiring high-speed training and inference
Projects needing advanced neural network architectures like convolutional or recurrent layers
Applications with large-scale datasets where performance and scalability are critical
Teams seeking a library with active maintenance, extensive documentation, or commercial support

Pros & Cons

Pros

Pure Ruby, No Dependencies

Being 100% Ruby with no external libraries makes it easy to install and run on any Ruby environment, eliminating setup hassles for educational use.

Faster Rprop Training

Implements resilient backpropagation, which trains significantly faster than standard backpropagation methods, as noted in comparisons with ai4r in the README.

Clear Educational Examples

Includes scripts for classification (iris species), regression (car MPG), and OCR (handwritten digits), providing hands-on learning with real-world datasets.

Minimal and Readable Code

With only about 250 lines of code focused on clarity, the implementation is easy to understand and modify for experimenting with neural network fundamentals.

Cons

Slow Performance

Explicitly acknowledged as significantly slower than production-grade libraries like ruby-fann, which wraps optimized C code, making it impractical for real-time or large-scale tasks.

Limited Feature Set

As a minimal implementation, it lacks advanced features such as support for different activation functions, layer types, or regularization techniques found in full-fledged ML libraries.

Cumbersome Example Setup

The OCR example requires downloading and setting up the MNIST dataset separately, adding complexity compared to self-contained educational tools.

Frequently Asked Questions

Related Projects

ruby-fann

Ruby library for interfacing with FANN (Fast Artificial Neural Network)

Stars506

Forks42

Last commit2 years ago

tlearn-rb

Recurrent Neural Network library for Ruby

Stars97

Forks15

Last commit8 years ago

brains

A feedforward neural network implementation for JRuby

Stars60

Forks1

Last commit9 years ago

cerebrum

Artificial Neural Networks (ANN's) in Ruby (unmaintained, unfortunately)

Stars36

Forks10

Last commit7 years ago

Community-curated · Updated weekly · 100% open source

Found a gem we're missing?

Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.

Submit a project Star on GitHub