neural network performance graph

Hi,

Could you please help me?

I had created a neural network by Matlab ANN toolbox and my network obtained MSE of 0.008 . I saved the workspace containing the network and variables. Now after some months i need to see its performance graph, the mse versus the epoch for my previously trained network. What command would help me have such thing?

Answer 1

[Greg Heath]

You can also recreate the training graph if you have saved the training history tr obtained from

[net tr ]= train(net,x,t);

Type

tr = tr

for details.

Hope this helps.

Thank you for formally accepting my answer.

Greg

Answer 2

[Greg Heath]

MSE =0.008 means absolutely nothing untill it is normalized. The best normalization factor is the MSE you would obtain from a naive model with a constant output, regardless of input. The minimum MSE occurs when the constant outputs are the means of the target variables. The resulting MSE is the average biased variance

MSE00 = mean(var(target',1))

and the normalized MSE for more complex models is

NMSE = MSE/MSE00.

The only way to recreate the training graph is to have saved the initial state of the random number generator rand that created the initial weights and train all over again.

Or you could have saved the initial weights ...

Hope this helps.

Thank you for formally accepting my answer.

Greg