Having issues going from trainNetwork to trainnet

I've attached 3 files(see post below for latest version of these files):

• trainNetworkEXAMPLE - my original trainNetwork implementation
• trainnetEXAMPLE - the trainnet implementation
• example.csv - data file with predictors and targets

The codes for the two examples are identical, the difference is only in the formatting of the input matrices.

trainNetworkEXAMPLE works as expected.

trainnetEXAMPLE works but convergence of the solver is different and training stops earlier.

Both codes end with

Training stopped: Met validation criterion

What am I getting wrong?

[Cris LaPierre]

You have changed the question. It is better to ask a new question.

Answer 1

[Cris LaPierre]

You have a vector sequence, so your layout should be s-by-c matrices, where s and c are the numbers of time steps and channels (features) of the sequences, respectively.

Withouth knowing more about your data, it looks like you have a vector sequence containing 1028 timesteps and 4 channels. You should therefore use the same code for creating XTrain, XValidation, TTrain, and TValidation as in your trainNetwork example. See here.

% Predictor values
XTrain = (XStandardized(1:numTimeStepsTrain,:));
XValidation = XStandardized(end-numTimeStepsTest+1:end,:);
% Target values 
TTrain = (TStandardized(1:numTimeStepsTrain,:));
TValidation = TStandardized(end-numTimeStepsTest+1:end,:);

[psousa]

Thanks for the answer! I've followed your suggestion and simplified the inputs on both files. I've attached them to this message.

Do you have any comment regarding the difference in how the two codes run?

[Cris LaPierre]

I'd remove this unnecessary code in your trainnet example:

% Predictor values
XTrain = XTrain;
XValidation = XValidation;
% Target values 
TTrain = TTrain;
TValidation = TValidation;

[psousa]

I left that there so that codes are very similar when comparing side-by-side.

Did you run both codes and see the differences in results?

Thank you for your answers!