"Unrecognized function or variable 'iPadArray'." when using a trained network to predict test data

I manually constructed a network and trained with GPU acceleration. But when I use the trained network to predict, I keep getting an error that "Unrecognized function or variable 'iPadArray'".

The codes that can reproduce the error are as follows

clear;clc

rng(17);

%% create the network with Resnet structure

wm = 'glorot';

net = layerGraph;

startLayer = [imageInputLayer([11,256,1],'Normalization','none');

convolution2dLayer([1 40],64,'Padding','same','Stride',[1 1],'WeightsInitializer',wm);

batchNormalizationLayer;

reluLayer;

maxPooling2dLayer([1 2],'Stride',[1 2],'Padding','same')];

net = net.addLayers(startLayer);

resunit = [convolution2dLayer([2 16],128,'Padding','same','Stride',[1 2],'WeightsInitializer',wm);

batchNormalizationLayer;

reluLayer;

convolution2dLayer([2 10],128,'Padding','same','Stride',[1 1],'WeightsInitializer',wm);

batchNormalizationLayer;

additionLayer(2)];

resunit_co = [convolution2dLayer([1 1],128,'Padding','same','Stride',[1 2],'WeightsInitializer',wm);

batchNormalizationLayer];

net = net.addLayers(resunit);

net = net.addLayers(resunit_co);

net = net.connectLayers('maxpool','conv_1');

net = net.connectLayers('maxpool','conv_3');

net = net.connectLayers('batchnorm_3','addition/in2');

lastlayer = [reluLayer;

convolution2dLayer([3 3],128,'Padding','same','Stride',[2 2],'WeightsInitializer',wm);

batchNormalizationLayer;

reluLayer;

fullyConnectedLayer(256,'WeightsInitializer',wm);

batchNormalizationLayer;

reluLayer;

fullyConnectedLayer(5,'WeightsInitializer',wm);

softmaxLayer];

net = net.addLayers(lastlayer);

net = net.connectLayers('addition','relu_2');

net = net.addLayers(classificationLayer);

net = net.connectLayers('softmax','classoutput');

%% train the network with randon data with gpu

xdata = rand(11,256,1,1e3);

ydata = categorical(randi(5,[1,1e3]));

trainOpt = trainingOptions('sgdm', ...

'MaxEpochs',10, ...

'MiniBatchSize',64, ...

'Plots','training-progress', ...

'Shuffle','every-epoch', ...

'ExecutionEnvironment','gpu', ...

'InitialLearnRate',1e-2, ...

'BatchNormalizationStatistics','population');

[netf, info] = trainNetwork(xdata,ydata,net,trainOpt);

%% make prediction and reproduce the error

xtest = rand(11,256,1,200);

ytest = netf.predict(xtest);

Answer 1

[Sai Pavan]

Hi Wenyu,

I understand that you are trying to resolve the error "Unrecognized function or variable iPadArray" which arises when you try to get the predictions on the test set with a trained model.

The error message typically occurs when a variable or function is used before it has been defined or due to a typographical error. Please refer to the below page to learn various possible causes of this error:

https://www.mathworks.com/matlabcentral/answers/96005-why-do-i-get-the-error-unrecognized-function-or-variable

Please refer to the below documentation to learn how to resolve the “Unrecognized function or variable” error:

https://www.mathworks.com/help/matlab/matlab_prog/resolve-error-undefined-function-or-variable.html

Another suggestion to address the error would be to convert the input test data into a GPU-compatible format using the “gpuArray” function before making predictions with the trained model.

Please refer to the below documentation to learn more about “gpuArray” function:

https://www.mathworks.com/help/parallel-computing/gpuarray.html

Hope it helps.

Regards,

Sai Pavan