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

2 ビュー (過去 30 日間)
SC
SC 2019 年 11 月 27 日
回答済み: Sourav Bairagya 2019 年 12 月 10 日
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 件のコメント
SC
SC 2019 年 11 月 30 日
編集済み: SC 2019 年 11 月 30 日
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

サインインしてコメントする。

回答 (1 件)

Sourav Bairagya
Sourav Bairagya 2019 年 12 月 10 日
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.

製品


リリース

R2019b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by