This repository contains work I've done for my ATCS Neural Networks class.
Each input file for the executable of Network.cpp is expected to be in the same directory as the executable. Note that examples of each file can be found in the src/ folder in this repository.
The input configuration parameter file is expected to contain the following values separated by whitespace in the following order: the number of connectivity layers in the network, the number of input activation nodes, the number of nodes in the hidden layers in order (each separated by whitespace), the number of nodes in the output layer, the number of test cases in the truth table, the name of the truth table file, and a boolean that is 1 if the network is in training mode or 0 if the network is in running mode.
If the network is in training mode, the following values should be included after the above in the following order: the name of the file to which the weights should be written while training, the value of the learning rate, the value of the error threshold to use where the network will cease training if the largest test case error in a given iteration is less than that error threshold, the maximum number of iterations for training, and a boolean that is 1 if the network should use random values for the initial weight values or 0 if the network should pull initial weight values from a file.
If the network is either not in training mode or should pull initial weight values from a file, the name of the weights file should be included after the above values.
If the network is in running mode, the name of the input activation layer file should be included after the above values.
Finally, if the network is in training mode and should use random values for the initial weight values, the following numbers should be included after the above in the following order: the minimum bound of the random range to use for weights and the maximum bound of the random range to use for weights.
The input truth table file is expected to contain numbers separated by whitespace. For each test case, the file should contain the values of the input activation nodes where the first value maps to the first input activation node, the second value maps to the second input activation node, and so on and so forth. The file should contain the values of the output layer nodes following these input activation node values that represent the expected output for the input activation values immediately before the output layer values. Note that the first output layer value maps to the first output layer node, the second output layer value maps to the second output layer node, etc. This pattern should then be repeated in the truth table file for each truth table test case.
The input weights file is expected to contain numbers separated by whitespace. To verify that the correct weights file is being used, the first numbers should be the following in the following order: the number of connectivity layers in the network, the number of input activation nodes, the number of nodes in the hidden layer, and the number of nodes in the output layer.
The following values should represent the weight values of the connections in the connectivity layers of the network. Each connectivity layer weight values should be in the order of the connections from the first activation node in the left layer to the nodes in the right layer, in order from the first to last nodes in the right layer, followed by the connections between the second activation node in the left layer to the nodes in the right layer, in the aforementioned order, and so on and so forth.
The input activation layer file is expected to contain the values of the input activation layer to use for running the network in order (where the first value corresponds to the first input activation node, the second value corresponds to the second input activation node, and so on and so forth) and separated by whitespace.