Does patternnet support single precision data?

I created a classic neural network using "patternnet". My GPU is optimized for single precision data, so how can I train the neural network using single precision data?

Answer 1

[MathWorks Support Team]

Unfortunately, single precision data is unsupported for "patternnet".

We advise you to use the new deep learning interface, which uses single precision by default. Please see below for code that creates a deep neural network with similar architecture to "patternnet" using the "iris" dataset in MATLAB:

[x,t] = iris_dataset;\nrng(0)\n[~, nData] = size(x);\n\n% Randomizing the data\nrandIDs = randperm(nData);\nXAllData = x(:,randIDs);\n\n% Data split for training and testing \nY = categorical(vec2ind(t)');\nXTrain = XAllData(:,1:100)';\nYTrain = Y(randIDs(1:100));\nXTest = XAllData(:,101:150)';\nYTest = Y(randIDs(101:150));\n\n% Defining the shallow neural network layers \nlayers = [featureInputLayer(4, "Name", "Input", "Normalization", "rescale-symmetric");...\n    fullyConnectedLayer(10, "Name", "firstHidden");\n    tanhLayer("Name","tanHyperbolicLayer");\n    fullyConnectedLayer(3, "Name", "OutputFC")\n    softmaxLayer\n    classificationLayer("Name","Output")];\n\n% Setting up training options\noptions = trainingOptions('sgdm', ...\n    'InitialLearnRate',1e-2, ...\n    'MaxEpochs',500,...\n    'Verbose',false, ...\n    "Shuffle","every-epoch", ...\n    "Plots","training-progress", ...\n    "ValidationData",{XTest,YTest});\n\n% Train the network\nnetDLT = trainNetwork(XTrain,YTrain,layers,options);

For more information about the available layers in the Deep Learning Toolbox, you may refer to the following documentation links:

https://www.mathworks.com/help/deeplearning/ug/list-of-deep-learning-layers.html

For more information about the "trainingOptions" and "trainNetwork" functions used in the above code, you may refer to the following documentation links:

https://www.mathworks.com/help/deeplearning/ref/trainingoptions.html

https://www.mathworks.com/help/deeplearning/ref/trainnetwork.html

To learn more about the new deep learning interface, you may refer to the following example:

​​​​​​​https://www.mathworks.com/help/deeplearning/gs/create-simple-deep-learning-classification-network.html