An Artificial Neural Network is computational implementation about 
    how the brain itself process the information. Where this computational
    implementation works in base of input(s)
    to have some output(s) as a result. Generally the goal 
    is to yield predictons that fits 
    the distribution of acceptable answers.

    The main components of an ANN with their respective purpose are the next ones:
        Node: Depending on the level of abstraction, the simplest node storage
            the result of the weighted sum that was generated by the predecessor nodes with their respective weights.
        Input Layer: It is a set of node(s) that has the main purpose to receive the
            input numeric data.
        Output Layer: It is a set of node(s) that has the main purpose to yield a numerical result,
            which is the prediction, following the forward propagation.
        Weight: It is a numerical variable that is in charge to be multiplied 
            with the data that pass trough to the node. With the main purpose to be changed. 
            If the prediction was not accurated we can manipulate the set of ANN's weights
            to have predictions which fit aceptable results.

    All the ANN learns by reducing the error. 
    The error is generated by the inaccuracy of the prediction in contrast
    with the real result. In order to make right predictions, we are able to adjust
    some variables inside of this network. This variables are called weights.
    In this case we are supervising the predictions that this ANN is yielding,
    and telling to the network how amount of error have predicted for each output.

    Making a prediction is basically take some numerical input(s)
    and follow the weighted sum of each layer's node until we reach the output layer.
    It is also called forward propagation because you are propaging the data 
    forward trough the network making some transormation 
    to this input data, due to the weights and the activation functions. 
    In order to obtain a suitable prediction.

    In order make the ANN more accurate
    we need to reduce the error that is producing.
    
    Firstly, we have to measure the error. 
    Nowadays exist plenty of ways to calculate this prediction's error.
    The basic idea is:
        error = prediction - real_result = delta
        
    Another ideas:
        error_abs = abs(prediction - real_result) = abs(error)
        error_squared = (prediction - real_result)^2 = error^2
     
    Secondly, we have to be aware about the releationship of the error and the weight, because obvisly if we change the weight(s) we
    are going to obtain a different error value. Thus the question is, each weight, where it needs to be 'move' 
    in order to have the less error as posible. In this case we only have to multiply delta which is the error,
    with the derivate of the input*weight product with respect to weight. In this case this derivate is only the input itself. Thus
    we calculate the weight_delta = delta * input, whihc is basically how much the weight caused an inaccurate result.
    And therefore we are able to correct the weight