how can ı use "minibatch​predict(ne​t,XTest);" command on simulink?

I trained a LSTM network.

How can I use "scores = minibatchpredict(net,XTest);" and "YPred = predict(net, XTest);" commands on Simulink?

Answer 1

[AJ Ibraheem]

The 'Stateful Predict' block might be what you're looking for. See https://uk.mathworks.com/help/deeplearning/ref/statefulpredict.html

[Bahadir]

Yes, I know the Stateful Predict. The results of Stateful Predict is different from minibatchpredict. Therefore, I want use minibatchpredict command on simulink. When I use minibatchpredict in the matlab function block, I get Error: Deep learning code generation using MinGW64 Compiler (C++) toolchain is not supported for mkldnn target. Function 'MATLAB Function' (#426.60.106).

I checked microsoft web site for Microsoft Visual C++ 2022, 2019, or 2017. There is Microsoft Visual C++ 2015-2022 Redistributable (x64) -14.42.34438.

[AJ Ibraheem]

> The results of Stateful Predict is different from minibatchpredict.

Yes it is understandable that you're getting different results. The stateful predict block would update the network states after prediction, to get the same result with the predict block, you'd have to follow [YPred, net.State] = predict(net, XTest) https://uk.mathworks.com/help/deeplearning/ref/dlnetwork.predict.html#d126e73498

> Using a MATLAB function block (as opposed to the Stateful predict block) for neural network prediction is valid.

It is possible that a layer in your network is not supported for the mkldnn target.

Do you mind sharing more about your current setup?

[Walter Roberson]

Using a Redistributable is not enough: you need to use the actual Visual C++ compiler. https://visualstudio.microsoft.com/thank-you-downloading-visual-studio/?sku=Community&channel=Release&version=VS2022&source=VSLandingPage&cid=2030&passive=false

[Bahadir]

@AJ Ibraheem

Thank you for your answer.

could you give more detail information about how to get same result on simulink. How to use predict command at matlab function block on simulink.

function y= fnc(u)
persistent net
if isempty(net)
    net = coder.loadDeepLearningNetwork('32.mat');
end
input= [u];
input=rescale(input); 
XTrain = {input'};  
output= predict(net, XTrain);
y=output{1};
end