Documented valid name-value pair argument is not recognised as parameter, for sequenceFolding and unfolding -layers

Hello,

I am trying to build a convolutional NN to classify time series data, and later combine it with an LSTM (similar to this study: LSTM Fully Convolutional Networks for Time Series Classification, Karim et al.)

To be able to use convolutional layer on time series MatLab gave me an error and suggested i use a sequenceFoldingLayer and sequenceUnfoldingLayer before and after the convolution layer. I Have not included both layers but get the following error:

Error using sequenceFoldingLayer>iParseInputArguments (line 74)
'NumOutputs' is not a recognized parameter. For a list of valid name-value pair arguments, see the documentation for this
function.

However, as stated in the documentation, 'NumOutputs' is a valid name-pair argument, and I do not understand why I cannot use it. If I try to not set 'NumOutputs' I get an error of unused output and input from the folding and unfolding layers because of a mismatch between the miniBatchSizes.

I have the following layers for the network setup:

inputSize = 1;

nClasses = 2;
miniBatchSize = 10;
layers = [...
    sequenceInputLayer(inputSize,'Name','InputLayer')

    sequenceFoldingLayer('NumOutputs', miniBatchSize,'Name', 'fold')
    convolution2dLayer(5,10)
    batchNormalizationLayer('Name', 'BN')
    reluLayer('Name','ReLu')
    sequenceUnfoldingLayer('NumInputs', miniBatchSize,'Name','unfold')
    
    fullyConnectedLayer(nClasses)
    softmaxLayer
    classificationLayer];

Answer 1

[Steven Lord]

The NumOutputs property is a documented property of custom layers as of release R2019a. The online documentation is for the most recent release, which is currently release R2019a. You're using an older release.

[Dams]

I am running MatLab R2019a. I get the following typing 'ver' in the Command Window:

-----------------------------------------------------------------------------------------------------
MATLAB Version: 9.6.0.1072779 (R2019a)
MATLAB License Number: <removed>
Operating System: Mac OS X  Version: 10.12.6 Build: 16G1314 
Java Version: Java 1.8.0_181-b13 with Oracle Corporation Java HotSpot(TM) 64-Bit Server VM mixed mode
-----------------------------------------------------------------------------------------------------
MATLAB                                                Version 9.6         (R2019a)
Simulink                                              Version 9.3         (R2019a)
Audio Toolbox                                         Version 2.0         (R2019a)
Computer Vision Toolbox                               Version 9.0         (R2019a)
Control System Toolbox                                Version 10.6        (R2019a)
DSP System Toolbox                                    Version 9.8         (R2019a)
Deep Learning Toolbox                                 Version 12.1        (R2019a)
Image Processing Toolbox                              Version 10.4        (R2019a)
Instrument Control Toolbox                            Version 4.0         (R2019a)
Optimization Toolbox                                  Version 8.3         (R2019a)
Signal Processing Toolbox                             Version 8.2         (R2019a)
Simulink Control Design                               Version 5.3         (R2019a)
Statistics and Machine Learning Toolbox               Version 11.5        (R2019a)
Symbolic Math Toolbox                                 Version 8.3         (R2019a)
>> 

[SL: edited to remove license number]

[Dams]

I have just made a clean install of MatLab R2019a and the error still persists.

[Steven Lord]

I don't think the layers you're using, sequenceFoldingLayer and sequenceUnfoldingLayer, are custom layers. I could be wrong but my understanding, based on the documentation pages, is that the only property of those layers you can modify is Name.

[Dams]

Yes, I see. That could be the issue. Though, I do not understand why the properties 'NumOutputs' and 'NumInputs' in the folding and unfolding layers respectively, are listed as properties that can be modified.

Thank you for your answer. I will try to use custom layers instead.

[Steven Lord]

Those layers have those properties, but I suspect they're read-only for those particular layers.

[Dams]

I finally found a solution. My problem was that I did not understand the documentation for the sequenceFolding and sequenceUnfolding layers, and missed a small, but very important fact, to use the connectLayers function to make the connection between the folding layers miniBatchSize output to the unfolding layers miniBatchSize input. The fact was hidden (in plain sight) in one line of text in the middle of an example.

I accept this answer because it lead me on the right way to understand.