Neural network

Neural network for multiclass classification created from scratch, built in Octave. For θ parameters optimization I have used fmincg function.

Getting Started

Prerequisites

You need to install Octave or Matlab to use this software.

Examples

% load X, y variables;

model = struct(); - you need to create struct where layers and parameters will be stored

% addLayer(model, number of units, activation function reference)

model = addLayer(model, size(X, 2), ''); - input layer (for input layer you do not specify activation function)
model = addLayer(model, 25, @sigmoid); - hidden layer
model = addLayer(model, 25, @sigmoid); - hidden layer
model = addLayer(model, 2, @sigmoid); - output layer

[model, cost] = trainNeuralNetwork(model, Xtrain, ytrain, lambda, number_of_iterations);

predictions = predict(model, Xtest, ytest);

Currently implemented activation functions

• Sigmoid

Authors

• Adam Bublavý - Initial work - Sangalaa

Licence

This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.