Undefined function or variable error using Neural Network Toolbox

I am following the provided example for Long-Short-Term-Memory networks found in the Neural Network Toolbox (https://de.mathworks.com/help/nnet/examples/classify-sequence-data-using-lstm-networks.html) and whenever I use the code given in the example for training the network

net = trainNetwork(X,Y,layers,options);

I get the following error:

>> net = trainNetwork(X,Y,layers,options);
Training on single GPU.
|=========================================================================================|
|     Epoch    |   Iteration  | Time Elapsed |  Mini-batch  |  Mini-batch  | Base Learning|
|              |              |  (seconds)   |     Loss     |   Accuracy   |     Rate     |
|=========================================================================================|
Error using trainNetwork (line 140)
Undefined function or variable 'endfunction'.

Caused by:
    Undefined function or variable 'endfunction'.

I tried re-installing the toolbox and rehasing the toolbox cache but it does not matter. Also, if I try to use

maxEpochs = 150;
miniBatchSize = 27;
shuffle = 'never';
cpu = 'cpu';

options = trainingOptions('sgdm', ...
    'MaxEpochs',maxEpochs, ...
    'MiniBatchSize',miniBatchSize, ...
    'Shuffle', shuffle), 'ExecutionEnvironment' , cpu);

I get this error message:

>> net = trainNetwork(X,Y,layers,options);
Error using trainNetwork (line 140)
Undefined function or variable 'endfunction'.

Caused by:
    Undefined function or variable 'endfunction'.

It looks like while using the CPU for computation the function does not even start.

Thanks in advance.

Answer 1

[Walter Roberson]

endfunction is for Octave code not MATLAB

[Walter Roberson]

It is possible that you have an octave function on your path that happens to have the same name as one of the Mathworks functions and so you could accidentally be invoking the wrong code.

To track this down you would

dbstop if caught error

and then run trainNetwork . You might have to dbcont a couple of times to get to the place where endfunction is being encountered.

[Burak Özbagci]

Thank you that was it. I removed any conflicting Octave codes found in my path. Now it works.

[Akib Rahman]

@Walter I am facing same type problem! I also use your suggestion ( dbstop if caught error ) but it is same. What should I do? In my PC neural network toolbox already installed.

When I run my code it gives me this error

Undefined function or variable 'batchNormalizationLayer'.

Error in leafRNN (line 30)
  batchNormalizationLayer

[Walter Roberson]

batchNormalizationLayer needs R2017b or later.

Answer 2

[Shounak Mitra]

options = trainingOptions('sgdm', ...
    'MaxEpochs',maxEpochs, ...
    'MiniBatchSize',miniBatchSize, ...
    'Shuffle', shuffle), 'ExecutionEnvironment' , cpu);

You might want to remove the bracket after shuffle? ;-)