Neural network coded entirely from scratch in C++
Project for Harker's ATCS: Neural Networks class

PART 1 - Execution Instructions

To run (requires makefiles): 

Open command line
Run "make" command (makefile is included)
Run ./output inputfile (configs)
where inputfile denotes the name/filepath of the input file and configs is the path
Of the config file containing hyper parameters (configs is optional and thus in parentheses)


PART 2 - Table of Contents

1. Main driver file (main.cpp)
Services: main() driver method, train() helper method

int train(int nOut, Network &n, int numIterations, double** trainData, double** truthVals)
   - Trains the given network using all the parameters fed into the network. 
   - Quits when the max iterations is reached or the network error goes below the 
     defined threshold. 
   - Returns 1 for successful train and 0 for unsuccessful train (max iterations reached 
     without going below error threshold)


2. Reader class (declared in reader.hpp and defined in reader.cpp)
Overall purpose: Reading in values and parameters from files, 
                 forming them into data structures and 
                 preparing them for network construction

Services: 

void readConfigFile(string config)
   - Reads a properly formatted config file containing
     hyperparameters such as lambda, max iterations, min error, and weight range. 
   - Stores these values in the hyperparameter global variables

void readMetaData(ifstream& fileIn)
   - Reads in the metadata of the network including
     number of training sets, layers of network, and
     whether the input has weights or not 
   - Stored in Reader object instance variables

void readTrainingData(ifstream& fileIn)
   - Reads in training data with the amount
     determined by the number of input activations
     given by network parameters and the metadata of
     number of training sets

void readWeights(ifstream& fileIn)
   - If the user requests for their own weights to be
     read in to the network, this method reads in the 
     weights from the file and properly formats them into
     the weights array. 


3. Network class (declared in network.hpp and defined in network.cpp)
Overall purpose: Containing the network constructs including forward
                 propagation and backpropagation for training. The network
                 services work for a generalized number/shape of layers

Services: 
double* run(double inputVals[])
   - Runs forward propagation through
     the network and returns the output layer
     aka the network outputs for the current input

void updateWeights()
   - Increments the weights using the backpropagation
     algorithm. The error is calculated *prior* to this step. 

int error()
   - Calculates the error after a certain output layer has been
     found by propagating input activations through the network.