To run this neural network, run Main.java

The program will create a neural network with the specifications contained in the configuration file. It will then randomize the weights of the network and train it using the data in the training data file. The output of the network will either be printed or be put into a bmp file specified by the user. The default output file is images/output.bmp. You can override the default file paths if you wish. Your files should have the following structures:

Structure of the Weights File

The default filename for the weights file is weights.txt. The format of this file is as follows: first, the size of each layer is given. Then, for each layer, the matrix for the weights is given. This matrix is such that the number of rows is the number of nodes in the previous activation layer, and the number of columns is the number of nodes in the next activation layer. Each column is space separated and each row is on the next line. There is a blank line between each layer and this format is repeated for each layer.

An example of a weights file would be:

2 2 1

0.5 0.5
0.5 0.5

0.3
0.3

Here, there are two connectivity layers, shown by the two matrices. The first layer connects two nodes to two nodes, and the second layer connects two nodes to one node.

Structure of the Image Training File

The default filename for the image training file is trainingImageFiles.txt. This file contains the filenames of all input and expected bmp files. These files will then be converted to an array and passed through the network. The structure of the file is as follows: The first line has the number of training cases, the height of each image and the width of each image. Then, the next lines contain the input file and the expected output file.

An example of a image training file is:

2 100 100
input1.bmp
expected1.bmp
input2.bmp
expected2.bmp

Here, there are two training cases, with input and expected images that have a total length of 100 each.

Structure of the Training Data File

The default filename for the training file is trainingData.txt. The format of this file is as follows: On the first line, the number of test cases is given. Then, on the following lines, for each test case, first the input values are given, space-separated, then the expected output values are given space-separated.

An example of a training data file is:

4 2 1
0 0
0
0 1
1
1 0
1
1 1
0

In this case, there are 4 test cases, each with 2 input nodes and 1 output node.

Structure of the Configuration File

The default filename for the config file is config.txt. The format of this file is as follows:

First, the size of each layer is given. Each of these sizes are given space-separated.

Each of the next lines contain a variable that configures a part of the training:

Min Weight:         the smallest value that the weights can be randomized to
Max Weight:         the largest value that the weights can be randomized to
Learning Rate:      the initial learning rate of the network
Lambda Multiplier:  how much to multiply the learning rate by each epoch
Maximum Epochs:     the number of epochs that will be run when training the network
Maximum Iterations: the number of times the network is retrained
Error Threshold:    the neural net stops when it goes below this error
Printing Rate:      how often to print the error during training