JavaNeuralNetworks

Playing around with Neural Networks.

Background

I saw Michael Nielson's book on Neural Networks mentioned on Hacker News, and read through the handwritten digits portion. I thought it'd be fun to see if I could recreate the Neural Network code as much as possible by hand. To that end, I created a Matrix class to deal with the Linear Algebra, and a NeuralNetwork class to handle the... well, Neural Network. It is all written in Java, and I hope to make a C++ version in the near future so that I can compare the performance.

Program Notes

The Neural Network works, but its a work in progress. Next steps are to clean up the source code, write unit tests, and see if I can't speed up the neural network.

Also, I'm not getting as good results as Nielson's program, so need to take another look at that.

Acknowledgements

This is based on Michael Nielsen's book "Neural Networks and Deep Learning" - Using neural nets to recognize handwritten digits. Available at Nielsen's website: http://neuralnetworksanddeeplearning.com/

Prerequisites

Download and extract the four files for training data available at http://yann.lecun.com/exdb/mnist/

To Compile

Edit the main file and provide the paths (in the trainNeuralNetwork method) for the training data downloaded above.
Path trainingImagesPath = Paths.get("path_to_train-images.idx3-ubyte");
Path trainingLabelsPath = Paths.get("path_to_train-labels.idx1-ubyte");
Path testLabelsPath = Paths.get("path_to_test-t10k-labels.idx1-ubyte");
Path testImagesPath = Paths.get("path_to_test-t10k-images.idx3-ubyte");
Open up a terminal/command line and navigate to src directory.
Enter the command
Windows: javac com\vafilor\neural\*.java
Linuxy: javac com/vafilor/neural/*.java

To Run

Open up a terminal/command line and navigate to src directory.
Enter the command: java com.vafilor.neural.Main