How can I get a confusion matrix after I train my AI using DeepNetworkDesigner?

I have trained an AI using deepnetwork designer app but i dont know how to create the confusion matrix

Answer 1

[Ganesh Gudipati]

Hi,

The confusion matrix can be built using the MATLAB function plotconfusion.

For more information refer plotconfusion.

[Powel Hancy]

Thank you for the replay

I have tried plotconfusion, but since I use the deep network designer app to train the AI when the training is finished I am only getting two variables in my work space 1.trainedNetwork_1 and 2.infostruct_1 I dont know which variables to use in the plotconfusion

and when I try plotconfusion in the command window it says variables not found.

[Ganesh Gudipati]

Hi,

Take some test data and test the trained network using predict function. Now using this results you can plot a confusion matrix.

Hope it helps.

[Powel Hancy]

Create and Train a Deep Learning Model

Script for creating and training a deep learning network with the following properties:

Number of layers: 25

Number of connections: 24

Training setup file: C:\Users\ph20aap\OneDrive - University of Hertfordshire\matlab_project_saves\faces_0001_64_gpu\faces_0001_64_gpu.mat

Run this script to create the network layers, import training and validation data, and train the network. The network layers are stored in the workspace variable layers. The trained network is stored in the workspace variable net.

To learn more, see Generate MATLAB Code From Deep Network Designer.

Auto-generated by MATLAB on 11-Apr-2022 19:53:23

Load Initial Parameters

Load parameters for network initialization. For transfer learning, the network initialization parameters are the parameters of the initial pretrained network.

trainingSetup = load("C:\Users\ph20aap\OneDrive - University of Hertfordshire\matlab_project_saves\faces_0001_64_gpu\faces_0001_64_gpu.mat");

Import Data

Import training and validation data.

imdsTrain = imageDatastore("C:\Users\ph20aap\OneDrive - University of Hertfordshire\Data sets\FACES\Faces","IncludeSubfolders",true,"LabelSource","foldernames");

[imdsTrain, imdsValidation] = splitEachLabel(imdsTrain,0.75,"randomized");

% Resize the images to match the network input layer.

augimdsTrain = augmentedImageDatastore([227 227 3],imdsTrain);

augimdsValidation = augmentedImageDatastore([227 227 3],imdsValidation);

Set Training Options

Specify options to use when training.

opts = trainingOptions("sgdm",...

"ExecutionEnvironment","gpu",...

"InitialLearnRate",0.0001,...

"MaxEpochs",50,...

"MiniBatchSize",64,...

"Shuffle","every-epoch",...

"ValidationFrequency",30,...

"Plots","training-progress",...

"ValidationData",augimdsValidation);

Create Array of Layers

layers = [

imageInputLayer([227 227 3],"Name","data","Mean",trainingSetup.data.Mean)

convolution2dLayer([11 11],96,"Name","conv1","BiasLearnRateFactor",2,"Stride",[4 4],"Bias",trainingSetup.conv1.Bias,"Weights",trainingSetup.conv1.Weights)

reluLayer("Name","relu1")

crossChannelNormalizationLayer(5,"Name","norm1","K",1)

maxPooling2dLayer([3 3],"Name","pool1","Stride",[2 2])

groupedConvolution2dLayer([5 5],128,2,"Name","conv2","BiasLearnRateFactor",2,"Padding",[2 2 2 2],"Bias",trainingSetup.conv2.Bias,"Weights",trainingSetup.conv2.Weights)

reluLayer("Name","relu2")

crossChannelNormalizationLayer(5,"Name","norm2","K",1)

maxPooling2dLayer([3 3],"Name","pool2","Stride",[2 2])

convolution2dLayer([3 3],384,"Name","conv3","BiasLearnRateFactor",2,"Padding",[1 1 1 1],"Bias",trainingSetup.conv3.Bias,"Weights",trainingSetup.conv3.Weights)

reluLayer("Name","relu3")

groupedConvolution2dLayer([3 3],192,2,"Name","conv4","BiasLearnRateFactor",2,"Padding",[1 1 1 1],"Bias",trainingSetup.conv4.Bias,"Weights",trainingSetup.conv4.Weights)

reluLayer("Name","relu4")

groupedConvolution2dLayer([3 3],128,2,"Name","conv5","BiasLearnRateFactor",2,"Padding",[1 1 1 1],"Bias",trainingSetup.conv5.Bias,"Weights",trainingSetup.conv5.Weights)

reluLayer("Name","relu5")

maxPooling2dLayer([3 3],"Name","pool5","Stride",[2 2])

fullyConnectedLayer(4096,"Name","fc6","BiasLearnRateFactor",2,"Bias",trainingSetup.fc6.Bias,"Weights",trainingSetup.fc6.Weights)

reluLayer("Name","relu6")

dropoutLayer(0.5,"Name","drop6")

fullyConnectedLayer(4096,"Name","fc7","BiasLearnRateFactor",2,"Bias",trainingSetup.fc7.Bias,"Weights",trainingSetup.fc7.Weights)

reluLayer("Name","relu7")

dropoutLayer(0.5,"Name","drop7")

fullyConnectedLayer(6,"Name","fc")

softmaxLayer("Name","prob")

classificationLayer("Name","classoutput")];

Train Network

Train the network using the specified options and training data.

[net, traininfo] = trainNetwork(augimdsTrain,layers,opts);

this is the script i genarated using DeepNetworkDesigner, I tried y=predict(net,augimdsValidation);

Unrecognized function or variable 'net'.

and it is giving me an error as I am new to matlab I am really struggiling to figure out the issue here.

[Ganesh Gudipati]

Export the network trained in Deep Network Designer to Simulink, on the Training tab, click Export > Export to Simulink.

Deep Network Designer saves the trained network as a MAT-file and generates Simulink blocks representing the trained network. The blocks generated depend on the type of network trained.

Now coonect the predict block from simulink library. Hope this solved the issue.

[Powel Hancy]

I am usining the matlab version R2021a as my university is not upgraded to the 2022 version so I will have to use the 2021a.... In 2021a there is only 2 options in export that is 1. Export to work space 2.Generate Live script. The genarated live script is the one I have posted above. Is there any other way to get the Confusion matrix from the Trained networks i have?

[Powel Hancy]

As I could not solve the issue in DeepNetworkDesigner, I found a temperory solution, that is create the network using DND and export the code before training it, and copy paste the code in Experiment manager app and do the training there.. Experiment manage will produce confusion matrix for you.