MATLAB Answers

How to change input values for weight classfication layer.

28 views (last 30 days)
Raza Ali
Raza Ali on 7 Oct 2019
Commented: Raza Ali on 14 Oct 2019
I am using weigth classfication fucntion which given as example in MATALAB documentaion.
But whenI use it in my network it gives error "Error using 'backwardLoss' in Layer weightedClassificationLayer. The function threw an error and could not be executed". I think the error is due to input value but i am not sure where to change these valuse. The weighted classification function works well according to input valuse assigned in example.
the code I am using for weighted classification function
%%%%%%
classdef weightedClassificationLayer < nnet.layer.ClassificationLayer
properties
% Row vector of weights corresponding to the classes in the
% training data.
ClassWeights
end
methods
function layer = weightedClassificationLayer(classWeights, name)
% layer = weightedClassificationLayer(classWeights) creates a
% weighted cross entropy loss layer. classWeights is a row
% vector of weights corresponding to the classes in the order
% that they appear in the training data.
%
% layer = weightedClassificationLayer(classWeights, name)
% additionally specifies the layer name.
% Set class weights.
layer.ClassWeights = classWeights;
% Set layer name.
if nargin == 2
layer.Name = name;
end
% Set layer description
layer.Description = 'Weighted cross entropy';
end
function loss = forwardLoss(layer, Y, T)
% loss = forwardLoss(layer, Y, T) returns the weighted cross
% entropy loss between the predictions Y and the training
% targets T.
N = size(Y,4);
Y = squeeze(Y);
T = squeeze(T);
W = layer.ClassWeights;
loss = -sum(W*(T.*log(Y)))/N;
end
function dLdY = backwardLoss(layer, Y, T)
% dLdX = backwardLoss(layer, Y, T) returns the derivatives of
% the weighted cross entropy loss with respect to the
% predictions Y.
[~,~,K,N] = size(Y);
Y = squeeze(Y);
T = squeeze(T);
W = layer.ClassWeights;
dLdY = -(W'.*T./Y)/N;
dLdY = reshape(dLdY,[1 1 K N]);
end
end
end

  0 Comments

Sign in to comment.

Accepted Answer

Pujitha Narra
Pujitha Narra on 11 Oct 2019
This is a way to initialize 'classWeights'
classWeights = 1./countcats(YTrain);
classWeights = classWeights'/mean(classWeights);
and you can use it here:
Network = [
imageInputLayer([256 256 3],"Name","imageinput")
convolution2dLayer([3 3],2,"Name","conv","Padding","same")
reluLayer("Name","relu")
softmaxLayer("Name","softmax")
weightedClassificationLayer(classWeights)
];
I think this should solve the problem.

  6 Comments

Show 3 older comments
Raza Ali
Raza Ali on 14 Oct 2019
Thank you for your help and sharing this link. I have been to this link, but my problem is still un solved.
I tried different valuse of Class weigths but getting same error again and again,
Let me explain you through example.
I have two classes A and B. both folders has 2500 images of 256 by 256 by 3 (in total 5000 images in two different classes).
I have tried following classweights.
  1. classWeights = rand(1,2)
  2. classWeights = rand(1,2500)
  3. classWeights = rand(1,5000)
  4. classWeights = rand(1,196608) % 256*256*3=196608
but no vlaues of classweight is working... Error using 'backwardLoss' in Layer weightedClassificationLayer. The function threw an error and could not be executed."
Raza Ali
Raza Ali on 14 Oct 2019
I am using two different image types( two classes A and B). Each Image has size: 256 by 256 by 3
%%%Start
imds = imageDatastore('Images','IncludeSubfolders',true,'LabelSource','foldernames');
[imdsTrain,imdsValidation] = splitEachLabel(imds,0.7,'randomized');
YTrain=imdsTrain.Labels;
YTrain = removecats(YTrain);
classWeights = 1./countcats(YTrain)
classWeights = classWeights'/mean(classWeights)
Network = [
imageInputLayer([256 256 3],"Name","data")
convolution2dLayer([3 3],16,"Name","conv1","BiasLearnRateFactor",2,"Stride",[4 4])
reluLayer("Name","relu1")
crossChannelNormalizationLayer(5,"Name","norm1","K",1)
maxPooling2dLayer([3 3],"Name","pool1","Stride",[2 2])
convolution2dLayer([3 3],32,"Name","conv","Padding","same")
reluLayer("Name","relu5")
maxPooling2dLayer([3 3],"Name","pool5","Stride",[2 2])
fullyConnectedLayer(2,"Name","fc8","BiasLearnRateFactor",2)
softmaxLayer("Name","prob")
weightedClassificationLayer("classWeights")
];
Options = trainingOptions('sgdm', ...
'MiniBatchSize',5, ...
'MaxEpochs',3, ...
'Shuffle','every-epoch', ...
'InitialLearnRate',1e-4, ...
'ValidationData',imdsValidation, ...
'ValidationFrequency',2100, ...
'Verbose',true, ...
'Plots','training-progress');
TrainedNetwork = trainNetwork(imdsTrain,Network,Options);

Sign in to comment.

More Answers (1)

Pujitha Narra
Pujitha Narra on 10 Oct 2019
Hi Raza Ali,
Can you mention how are you using 'weightedClassificationLayer' in your network? Assuming you want to know the inputs to the constructor of this class:
'classWeights' and the layer's 'name' are the only inputs.
'classWeights'-. classWeights is a row vector of weights corresponding to the classes in the order that they appear in the training data.
'name' -additionally specifies the layer name.
Also this example might be of help
Hope this helps!

  7 Comments

Show 4 older comments
Pujitha Narra
Pujitha Narra on 11 Oct 2019
Hi Raza,
What do you mean when you say, you want to take classWeights from softmax layer?
Raza Ali
Raza Ali on 11 Oct 2019
I just want to use weightClassfication layer in simple CNN layer as output layer. and my image size is 256 x 256 x 3.
The nnetwork configurations i have mentioned in question. but when i use it gives error.
Error using 'backwardLoss' in Layer weightedClassificationLayer. The function threw an error and could not be executed.
Raza Ali
Raza Ali on 11 Oct 2019
Network = [
imageInputLayer([256 256 3],"Name","imageinput")
convolution2dLayer([3 3],2,"Name","conv","Padding","same")
reluLayer("Name","relu")
softmaxLayer("Name","softmax")
weightedClassificationLayer('classWeights')
];

Sign in to comment.

Community Treasure Hunt

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

Start Hunting!