train network error. reshape number of elements must not change. Use [] to automatically calculate the dimension size.

my input is a combined datastore of - 28x28 double input, and 1x1 categorical label.

layers = [

imageInputLayer([28 28 1],"Name","imageinput","Normalization","none")

fullyConnectedLayer(1,"Name","fc_2")

softmaxLayer("Name","softmax")

classificationLayer("Name","classoutput")];

miniBatchSize = 28;

options =

trainingOptions('adam', ...

'ExecutionEnvironment','cpu', ...

'MaxEpochs',84, ...

'MiniBatchSize',miniBatchSize, ...

'GradientThreshold',2, ...

'Shuffle','never',...

'Verbose',0, ...

'Plots','training-progress');

net = trainNetwork(cdsTrain,layers,options);

i get an error at trainnetwork script:

Error using trainNetwork (line 170)

Number of elements must not change. Use [] as one of the size inputs to automatically calculate the appropriate size for that dimension.

Caused by:

Error using reshape

Number of elements must not change. Use [] as one of the size inputs to automatically calculate the appropriate size for that dimension.

I have tried quite a few similar designs for layers, by adding a few more layers in different manners, but they all display the same error, so i suspect it has something to do with the fully connected layer.

Answer 1

[Mahesh Taparia]

Hi

You are having an input of size MxN and you are directly reshaping it to a vector of length 1x1. Before the fullyConnectedLayer, there is a need to vectorized the features. To do this you can use flattenLayer before the fullyConnectedLayer. For more information, you can refer to this documentation of flattenLayer.