How to do the sum for 2 gradient objects in the deep learning toolbox?

Question

SC 2019-11-27

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/493430-how-to-do-the-sum-for-2-gradient-objects-in-the-deep-learning-toolbox

回答： Sourav Bairagya 2019-12-10

Hi,

I have gradients1 and gradients2 which have exactly same structure but different numerical values. How can I do the sum? Current I tried gradients1+gradients2 but I got error.

Thanks!

My code:

rng(123); % seed
X_ori=[4,163,80;5,164,75]; % data; #(number) = 2; #(features) = 3;
X=permute(X_ori,[3,4,2,1]);
dlX = dlarray(X, 'SSCB');
Y_ori=[0, 0, 0, 1; 0, 1, 0, 0]; % data labels (i.e. one-hot vectors for 4 classes)
myModel = [
    imageInputLayer([1 1 3],'Normalization','none','Name','in')
    fullyConnectedLayer(7,'Name','Layer 1')
    fullyConnectedLayer(4,'Name','Layer 2')];
MyLGraph = layerGraph(myModel);
myDLnet = dlnetwork(MyLGraph);
gradients1 = dlfeval(@modelGradients1, myDLnet, dlX, Y_ori);
gradients2 = dlfeval(@modelGradients2, myDLnet, dlX, Y_ori);
gradients_sum = gradients1+gradients2;
function [gradients1] = modelGradients1(myModel, modelInput, CorrectLabels)
CorrectLabels_transpose=transpose(CorrectLabels);
[modelOutput,state] = forward(myModel,modelInput);
loss = -31*sum(sum(CorrectLabels_transpose.*log(sigmoid(modelOutput/100))));
gradients1 = dlgradient(loss, myModel.Learnables);
end
function [gradients2] = modelGradients2(myModel, modelInput, CorrectLabels)
CorrectLabels_transpose=transpose(CorrectLabels);
[modelOutput,state] = forward(myModel,modelInput);
loss = -42*sum(sum(CorrectLabels_transpose.*log(sigmoid(modelOutput/100))));
gradients2 = dlgradient(loss, myModel.Learnables);
end

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

SC 2019-11-30

编辑：SC 2019-11-30

在 MATLAB Online 中打开

I can change the line "gradients_sum = gradients1+gradients2;" to the followings and then I can get the sum. But I still want to know if there are some more efficient ways to do so.

gradients_sum = grad_sum(gradients1, gradients2);
function gradients_sum=grad_sum(gradients1, gradients2)
num_layers=size(gradients1,1);
gradients_sum=gradients1;
for i=1:num_layers
    gradients_sum.Value{i,1}=gradients1.Value{i,1}+gradients2.Value{i,1};
end
end

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Sourav Bairagya 2019-12-10

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/493430-how-to-do-the-sum-for-2-gradient-objects-in-the-deep-learning-toolbox#answer_405735

As in this case, 'gradients1.Value' and 'gradients2.Value' both are cell arrays and each one contains another cell arrays as elements within it, hence, direct conversion of these two cell arrays into matrices using 'cell2mat' or direct addition of them using '+' operator is not possible. Hence, you have to access each element individually and add them.