Why the final Validation accuracy appears on the plot different than the accuracy that is calculated by the law of accuracy ?

I create and train a simple convolutional neural network for deep learning classification on Matlab, when training finishes, the final validation accuracy that appears on the right side of the plot is different than the accuracy I have gotten from the following law for the validation set

accuracy = sum(predictedLabels == valLabels)/numel(valLabels);

[Maria Duarte Rosa]

Hi Nusaiba,

Which MATLAB version are you using?

Answer 1

[Manlin Wang]

I have the same problem. The validation accuracy showed in the training process plot is different from the law of accuracy.The codes are posted as follows:

indices=crossvalind('Kfold',length(XTrain),5);
validate_data_in = (indices == 1); 
train = ~validate_data_in;
Xvalidate_data=XTrain(validate_data_in,:);%测试集为20%数据
Yvalidate_label=YTrain(validate_data_in,:);%测试指标
Xtrain_data=XTrain(train,:);
YTrain_label=YTrain(train,:);
indices1=crossvalind('Kfold',length(Xtrain_data),10);
acc=zeros(2,1);
for k=1:2
test = (indices1 == k); 
train = ~test;
X_train=Xtrain_data(train,:);
Y_Trainlabel=YTrain_label(train,:);
test_data=Xtrain_data(test,:);
test_target=YTrain_label(test,:);
inputSize = 31;
numHiddenUnits = 120;
numClasses = 2;
layers = [ ...
    sequenceInputLayer(inputSize)
    bilstmLayer(numHiddenUnits,'OutputMode','sequence')
    dropoutLayer(0.2)
    bilstmLayer(100,'OutputMode','sequence')
    dropoutLayer(0.2)
    bilstmLayer(50,'OutputMode','last')
    dropoutLayer(0.2)    
    fullyConnectedLayer(numClasses)
    softmaxLayer
    classificationLayer];


% 
maxEpochs = 100;
miniBatchSize = 100;
% lgraph = layerGraph(layers);
% lgraph = connectLayers(lgraph,'fold/miniBatchSize','unfold/miniBatchSize');
options = trainingOptions('sgdm', ...
    'ExecutionEnvironment','cpu', ...
    'GradientThreshold',1, ...
    'MaxEpochs',maxEpochs, ...
    'MiniBatchSize',miniBatchSize, ...
    'InitialLearnRate',1e-3, ...
    'SequenceLength','longest', ...
    'ValidationData',{test_data,test_target}, ...
    'Shuffle','every-epoch', ...
    'Verbose',false, ...
    'Plots','training-progress');
%train LSTM network
net = trainNetwork(X_train,Y_Trainlabel,layers,options);
%Test LSTM Network
YPred = classify(net,test_data);
acc(k) = sum(YPred == test_target)./numel(test_target)
end

[Maria Duarte Rosa]

Hi Manlin Wang,

Which MATLAB version are you using?

[Xinlong Liu]

Hi Maria,

I have the same problem. I am using MATLAB R2017b. Are there any solutions?