Error: Invalid network layer does not support sequence input

hi, i am building a CNN training model. however, i got this error that do not know how to solve. i tried to insert a sequence folding layer then i got error again saying that "unconnected input and output". please help me with this

% Load training data and essential parameters
load('trainData.mat','XTrain','YTrain');

numSC = 64;

% Batch size
miniBatchSize = 4000;

% Iteration
maxEpochs = 10;

% Sturcture
inputSize = [6,64,1];
numHiddenUnits = 128; 
numHiddenUnits2 = 64;
numHiddenUnits3 = numSC;
numClasses = 16;

% DNN Layers
layers = [ ...
    sequenceInputLayer(inputSize,'Name','sequence')
    convolution2dLayer(3,32,'Name','conv2')
    reluLayer('Name','relu')
    maxPooling2dLayer(2,'Name','maxpool')
    flattenLayer('Name','flat')
    lstmLayer(numHiddenUnits,'OutputMode','last','Name','lstm')
    fullyConnectedLayer(numClasses,'Name','fc')
    softmaxLayer('Name','sm')
    classificationLayer('Name','class')];

% Training options
options = trainingOptions('adam',...
    'InitialLearnRate',0.01,...
    'ExecutionEnvironment','auto', ...
    'GradientThreshold',1, ...
    'LearnRateDropFactor',0.1,...
    'MaxEpochs',maxEpochs, ...
    'MiniBatchSize',miniBatchSize, ...
    'Shuffle','every-epoch', ...
    'Verbose',1,...
    'Plots','training-progress'); 

% Train the neural network
tic;

net = trainNetwork(XTrain,YTrain,layers,options);

toc;

save('NN.mat','net');

Answer 1

[Mahesh Taparia]

Hi

There is a requirement of sequenceFoldingLayer and sequenceUnfoldingLayer in the layer graph. For a sample layergraph, you can refer here. You can consider the below code for your case:

% DNN Layers
layers = [ ...
    sequenceInputLayer(inputSize,'Name','sequence')
    sequenceFoldingLayer('Name','fold')
    convolution2dLayer(3,32,'Name','conv2')
    reluLayer('Name','relu')
    maxPooling2dLayer(2,'Name','maxpool')
    sequenceUnfoldingLayer('Name','unfold')
    flattenLayer('Name','flat')
    lstmLayer(numHiddenUnits,'OutputMode','last','Name','lstm')
    fullyConnectedLayer(numClasses,'Name','fc')
    softmaxLayer('Name','sm')
    classificationLayer('Name','class')];

lgraph = layerGraph(layers);
lgraph = connectLayers(lgraph,'fold/miniBatchSize','unfold/miniBatchSize');
analyzeNetwork(lgraph)

Hope it will help!

[ZEMIN HUANG]

Hi!! it works for me, thank you!!!