Neural Networks

TP2 for Artificial Intelligence Systems course from I.T.B.A.

Documentation

Project related documents can be found at the docs folder

Perceptrons

The project has included 6 perceptrons:

• Currently working:
  • TERRAIN PERCEPTRON
• Currently not working (used in development stage only):
  • AND, OR, PARITY, SYMMETRY, XOR PERCEPTRONS

Pre-requisites

Add the src folder & subfolders to the Matlab/Octave path.

Configuration

Change the src/get_config.m file with desired parameters.

Change non linear activation function's beta

case 'non_linear_exp_g'
    out.beta = 1/2;
case 'non_linear_tanh_g'
    out.beta = 1;'

Change adaptive learning rate parameters

Where a and b are the adaptive constants and k refers to the number of continuous good epochs that have to occur in order to increment the learning rate.

case 'evaluate_gap'
    out.a = 0.001;
    out.b = 0.1;
    out.k = 3;

Change when to plot while training

case 'neural_network'
    out.net_save_period = 50;
    out.error_bars_plot_period = 15;

Terrain perceptron

Change the filename where the trained network will be saved

case 'terrain_perceptron'
    out.filename = 'terrain_perceptron_net.mat';
    ...

Change the perceptron's data

Where data_size is the sample size to take from patterns and expected_outputs. terrain_data() function returns the provided data for the perceptron.

case 'terrain_perceptron'
    ...
    [out.patterns, out.expected_outputs] = terrain_data();
    out.data_size = columns(out.patterns);
    ...

Change the perceptron's parameters

• epsilon is the error difference that has to be reached in order to stop training when comparing the network's outputs with the desired ones for the training patterns.
• alpha is the momentum parameter.
  • Disable momentum: set alpha = 0
• gap.size is the number of epochs that need to have passed in order to test if a good epoch was reached.
  • IMPORTANT: Currently unavailable option: You should always use gap.size = 1
• gap.eval it's the adaptive learning rate function. Possible values:
  • Use adaptive learning rate: @dont_evaluate_gap
  • Disable adaptive learning rate: @evaluate_gap

 case 'terrain_perceptron'
    ...
    out.epsilon = 0.1;
    out.eta = 0.05;
    out.alpha = 0.9;
    out.gap.size = 1;
    out.gap.eval = @dont_evaluate_gap;
    ...

Change the perceptron's architecture

• layers.layers.neurons is an array where each element represents a layer and sets the number of neurons for that layer.
• layers.hidden.g is the activation function for all the hidden layers and layers.hidden.g_derivative its derivative.
• layers.last.g is the activation function for the last layer and layers.last.g_derivative its derivative.
• layers.*.g possible values:
  • non_linear_tanh_g
  • non_linear_exp_g
  • linear_identity_g
• layers.*.g_derivative possible values:
  • non_linear_tanh_g_derivative
  • non_linear_tanh_g_derivative_improved
  • non_linear_exp_g_derivative
  • non_linear_exp_g_derivative_improved
  • linear_identity_g_derivative

case 'terrain_perceptron'
    ...
    out.layers.neurons = [rows(out.patterns), 10, 5, 5, rows(out.expected_outputs)];
    out.layers.hidden.g = @non_linear_tanh_g;
    out.layers.hidden.g_derivative = @non_linear_tanh_g_derivative_improved;
    out.layers.last.g = @linear_identity_g;
    out.layers.last.g_derivative = @linear_identity_g_derivative;

Usage

Run the desired perceptron's script from src/perceptrons.

Terrain Perceptron

Run src/perceptrons/tarrain_perceptron/terrain_perceptron.m script after changing the perceptron's configuration with desired parameters.

The script will save a file with the name specified in the configuration with the trained network data. You can use this trained network to solve other data without learning by running the src/perceptrons/tarrain_perceptron/trained_terrain_perceptron.m script after changing it's content with desired parameters.

Example

From the Matlab/Octave command line, and after adding the src folder & subfolders to the Matlab/Octave path, execute

terrain_perceptron

Then, you can do

trained_terrain_perceptron

Authors

This project is written and maintained by

• Matías Nicolás Comercio Vázquez
• Gonzalo Ibars Ingman
• Matías Mercado
• Juan Moreno

License

MIT License

Copyright (c) 2017
  - Matías Nicolás Comercio Vázquez <mcomerciovazquez@gmail.com>
  - Gonzalo Ibars Ingman <gibarsin@itba.edu.ar>
  - Matías Mercado <mmercado@itba.edu.ar>
  - Juan Moreno <jpmrno@itba.edu.ar>

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.