Must optimize the learning algorithm

[sreramk]

the method

    public void computeNetworkResult(int startIndex, MoveOrder direction) throws Exception {

in the file NeuralNetwork.java, obviously has two arguments int startIndex and MoveOrder direction which respectively determine the layer (within the set of hidden layers) from which the evaluation process must start and the second argument defines how the process should proceed (forward or backwards). The functionality of the second argument is completely unnecessary and must be removed. The first argument only works as expected if the value zero is passed into it. This argument, when made to work, can be made to evaluate the result from the chosen layer. Without this, for each learning iteration, the entire system is computed for three complete times. With this feature, the total execution time of the above method can be cut by half.
This method (computeNetworkResult) being the slowest one in the method,

      // this method is present in the file: NeuralTrainer.java
    public void trainNeuron(Neuron neuron, ArrayList<Double> expectedOutput) throws Exception {

increases the overall training speed of the neural network system by twice the previous execution speed. Any possible optimization will be really helpful for this project.

[sreramk]

The optimization described here is added in the latest version. But this does not seem to improve the speed.

[sreramk]

I have switched over to the well accepted gradient-decent algorithm because my algorithm does not seem to work. The training algorithm is no longer an issue, but the part that talks about the evaluate method needs attention. I haven't tested it yet.
But it is safe to remove it from the first milestone.