Error using trainNetwork (line 184)

Hi All,

I am trying to classify sequence datasets using deep learning by applying

net = trainNetwork(XTraining,YTraining,layers,options);

where

K>> whos XTraining
  Name              Size               Bytes  Class    Attributes

  XTraining      6586x1             63910544  cell    
  
K>> whos YTraining
  Name              Size            Bytes  Class          Attributes

  YTraining      6586x1              9598  categorical  

but I am getting the following error

Error using trainNetwork (line 184)
Out of range subscript.

Caused by:
    Error using sub2ind (line 43)
    Out of range subscript.

I don't get why I get the out of range subscript error since XTraining & YTraining are both 6586x1.

However if I try

    net = trainNetwork(XTraining(1:4000),YTraining(1:4000),layers,options);

it works well but not for any ranges above 4000 :|

Can someone please let me know what's going on and how I can solve this issue?

Answer 1

[Walter Roberson]

What would happen if one of the layers requested an array larger than the size stored in your cells, XTraining{1} and so on?

In particular if it fails for values above 4000 check that XTraining{4001} is the same size as XTraining{1}

[Susan]

Thanks for your reply. I checked that out. All have the same size, i.e.,

size(XTraining{4001}) = size(XTraining{1})= size(XTraining{6586}) = 2x600

[Susan]

Any other suggestions?

[Walter Roberson]

Not at the moment. If you post your code and data then I might be able to get further, perhaps.

[Susan]

That would be great! Thanks in advance.

When I'm attaching the data I got the error "XTraining.zip size exceeds 5 MB". I've tried other format compressing the data file in an archive format such as tart. giz but these format are not acceptable. rather chucking data set, is there any other ways?

[Walter Roberson]

Google Drive ?

[Susan]

Sure thing!

Here is the code. Label data is attached and the traing data can be find here https://drive.google.com/file/d/1mJ4Fu9SkiYos2yOVykLssiABxuMeDlyA/view?usp=sharing

layers = [ ...
    sequenceInputLayer(2)
    bilstmLayer(100,'OutputMode','last')
    fullyConnectedLayer(19)
    softmaxLayer
    classificationLayer
    ];

options = trainingOptions('adam', ...
    'MaxEpochs',10, ...
    'MiniBatchSize', 50, ...
    'InitialLearnRate', 0.01, ...
    'SequenceLength', 1000, ...
    'GradientThreshold', 1, ...
    'ExecutionEnvironment',"auto",...
    'plots','training-progress', ...
    'Verbose',false);

net = trainNetwork(XTraining,YTraining,layers,options);

By the way, when I wan't to use the classification learner app by uploading these data from workspace, the Xtraining data doesn't show up. Any idea how I can apply the app to these data?

Many many thanks in advance!

[Susan]

I figured out what I did wrong! thanks for offering your help though!

Could you please help me with applying the classification learner app? I think the issue is the classification learner APP does not take cell array as input. How to deal with this issue?

Thanks in advance!

[Emma Jakobsson]

I solved it too

[Ratri M]

I am getting same problem too, would you mind telling what was your solution?

[Juan Leal]

Hi, could you please explain how you solved this issue? I am also having the same problem.

[Vinay Kulkarni]

Facing same issue. Can you please help on how you solved it.

[Susan]

@Vinay Kulkarni In my case, I had forgotten to change the number of layers in fullyConnectedLayer according to the number of classes. And there were some issues with the length of data in each cell, so I got to ensure that all data had the same size and length. Hope it helps.

[Vinay Kulkarni]

Thanks @Susan.

I too had made mistake in number of classes. Once that corrected the training started.