I am working on a project which should predict values based on past data. I need to implement the sliding window technique where the network training is based over a sequence of values.
I am able to record the error reported once the back propogation perform the learning, however I need a way of aquiring the error after loading unseen data into the network.
This is required to ensure that there is no overfitting whilst ensuring that the weights are set to the right values when predicting an output which input values were never seen by the network.
Would appreciate any assistance in this regards.