netTrained = trainnet(s​equences,t​argets,net​,lossFcn,o​ptions)，se​quences包含复​数无法使用此函数

问题：

应用函数netTrained = trainnet(sequences,targets,net,lossFcn,options)，

sequences包含复数时如何使用此函数？

函数说明里有提示可使用复数输入：This argument supports complex-valued predictors and targets.

代码：

XTrain = permute(dataTrain(:,1:end-1,:),[1,3,2]);

TTrain = permute(dataTrain(:,2:end,:),[1,3,2]);

numChannels = betalen;

layers = [

sequenceInputLayer(numChannels)

lstmLayer(128)

fullyConnectedLayer(numChannels)];

options = trainingOptions("adam", ...

MaxEpochs=200, ...

SequencePaddingDirection="left", ...

Shuffle="every-epoch", ...

Plots="training-progress", ...

Verbose=false);

net = trainnet(XTrain,TTrain,layers,"mse",options);

报错结果：

错误使用 trainnet (第 46 行)

在层 'lstm' 期间执行失败。

出错 HDL (第 66 行)

net = trainnet(XTrain,TTrain,layers,"mse",options);

原因:

错误使用 dlarray/lstm (第 105 行)

位置 1 处的参数无效。 值必须为实数。

Answer 1

[Paras Gupta]

Hi Alexander,

I understand that you are trying to use the "trainnet" function on complex-valued sequences and complex-valued targets.

You are correct in noting that the documentation indicates that the "trainnet" function can support complex-valued predictors and targets. However, the built-in loss functions provided by "trainnet" do not inherently support complex-valued targets. To address this, you will need to define a custom loss function that can handle complex values for targets.

Moreover, the "sequenceInputLayer" in your model should be configured to handle complex-valued inputs. This can be done by setting the "SplitComplexInputs" argument to true.

Below is an example of a custom loss function for complex inputs, which you can use in your training loop:

% dummy data
numSamples = 100; 
numTimesteps = 10;
numChannels = 2; 
realPart = randn(numSamples, numTimesteps, numChannels);
imagPart = randn(numSamples, numTimesteps, numChannels);
dataTrain = realPart + 1i * imagPart;

XTrain = permute(dataTrain(:,1:end-1,:),[1,3,2]);

% complex target
TTrain = permute(dataTrain(:,2:end,:),[1,3,2]);
% real target
% TTrain = rand(numSamples, numChannels, numTimesteps-1);

numChannels = 2;
layers = [
    sequenceInputLayer(numChannels, SplitComplexInputs=true) % split Complex Inputs
    lstmLayer(128)
    fullyConnectedLayer(numChannels)];

options = trainingOptions("adam", ...
    MaxEpochs=200, ...
    SequencePaddingDirection="left", ...
    Shuffle="every-epoch", ...
    Plots="training-progress", ...
    Verbose=false);


% net = trainnet(XTrain, TTrain, layers, "mse", options);
% custom loss function passed as function handle
net = trainnet(XTrain, TTrain, layers, @complexLoss, options);

function loss = complexLoss(Y, T) 
    difference = Y - T;
    squaredMagnitude = real(difference).^2;
    loss = mean(squaredMagnitude, 'all');
end

You can refer to the following documentation links for more information on the code above:

• 'Train Network with Complex-Valued Data' example - https://www.mathworks.com/help/deeplearning/ug/train-network-with-complex-valued-data.html
• "SplitComplexInputs" argument - https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.sequenceinputlayer.html#mw_1f1ef68c-244a-4374-a846-2e24b71d384f_sep_mw_19bc7780-8e05-482a-b309-d24e230ab466
• function handles - https://www.mathworks.com/help/matlab/function-handles.html
• "trainnet" function - https://www.mathworks.com/help/deeplearning/ref/trainnet.html

Hope this helps with your work.