I am encountering a persistent error with the predict function in the Deep Learning Toolbox within MATLAB 2024b.

The problem occurs even in a simplified test script designed to isolate the issue:

Matlab

% Basic Script to Test dlarray and predict
function network = initializeSimpleActorNetwork(inputSize, outputSize)
layers = [
    featureInputLayer(inputSize, 'Name', 'state_input')
    fullyConnectedLayer(outputSize, 'Name', 'output')
    ];
network = dlnetwork(layerGraph(layers));
end
inputSize = 1827;
outputSize = 10;
testDlarray = dlarray(randn(1, inputSize), 'CB');
testNetwork = initializeSimpleActorNetwork(inputSize, outputSize);
try
    testPredict = predict(testNetwork, testDlarray);
    disp('Prediction successful!');
    disp(['Output size: ', num2str(size(testPredict))]);
catch ME
    disp('Prediction failed!');
    disp(['Error message: ', ME.message]);
end

Despite the dlarray being created with the correct size (1x1827) and the featureInputLayer being configured to accept an input of size 1827, the predict function consistently throws an error stating that it received an input of size 1.

Troubleshooting Steps Taken:

• Reinstalled MATLAB 2024b.
• Verified that the issue persists in the isolated test script.
• Confirmed that the dlarray is correctly created and that the input layer is correctly configured.

[Chika]

Error message: Layer "state_input": Invalid input data. Invalid size of channel dimension. Layer expects input with channel dimension size 1827 but received input with size 1.

Answer 1

[Matt J]

You need,

testDlarray = dlarray(randn(inputSize,1), 'CB');

or,

testDlarray = dlarray(randn(1, inputSize), 'BC');

[Chika]

this worked for the section of code . Although other errors are still there.

Thanks for your quick and brilliant response.