Customization of performance function for Neural Network Time-Series Prediction and Modeling in which the target is normalized by external time series values

Dear all,

We are struggling to figure out if it is possible to normalize the target for the performance function involving an external time series.

For instance, given the target t and the estimated output y we can easily obtain the loss function by using Matlab function:

perf = mse(net,t,y);

Now, we ask is it possible to define a new target such that (and therfore ), where x is a function of time that is not used for network training?

% for instance
x = rand(length(t)); 

y_new = y./x;
t_new = t./x;
my_perf = mse(net, t_new, y_new);

Unfortunately, when training the network we don't manage to provide to mse fucntion and .

Thank you in advance,

Marco and Francesco

Answer 1

[yanqi liu]

sir，may be it is not suitable，if the inputlayers has rescale，such as transfer to [0, 1]，may be the result can compute

[Marco Pizzoli]

Thank you for the reply.

However I don't get what you mean.

Regards,

Marco