yeah i got it, i guess this is iterative process of improvement just like gradient descent optimizes the weights to get the optima