Error loading MNIST Images

Hello,

I'm new to Matlab in general, and I'm trying to setup a multilayer perceptron, for the images of the MNIST Database classification. (Handwriting Recognition)

I'm using the loadMNISTImages function, to load the images to my network, but I get an error in nntrain function that my train_set must be a float.

How can I make it work? Here is my code:

clear;clc; format compact;

train_set = loadMNISTImages('train-images.idx3-ubyte', 10000);

train_label = loadMNISTLabels('train-labels.idx1-ubyte', 10000);

test_set = loadMNISTImages('t10k-images.idx3-ubyte', 1000);

test_label = loadMNISTLabels('t10k-labels.idx1-ubyte', 1000);

%train_set = train_set.reshape(train_set.shape[0], 28 ,28 ,1);

train_label = double(train_label);

%test_set = double(test_set)'/255;

test_label = double(test_label);

%%%%%%%%%%%%%%

% Train MLP %

%%%%%%%%%%%%%%

nhid = 40; %Nr. of hidden nodes

nn = nnsetup([784 14 10]); %Specify MLP architecture

nn.activation_function = 'tanh_opt'; %Set hidden neuron activstion function

nn.learningRate = 0.15; %set learning rate

%nn.learningRate = 1;

%nn.learningRate = 1.5;

nn.momentum = 0.9; %set momentum

nn.weightPenaltyL2 = 1e-4; %set regularization parameter λ

nn.dropoutFraction = 0.5;

nn.output = 'softmax';

opts.batchsize = 100; %specifies version of batch backpropagation

opts.numepochs = 100; %set number of epochs

opts.plot = 1; %Show plot of training error vs epochs

[nn, L] = nntrain(nn, train_set, train_label, opts);

[er1, bad] = nntest(nn, test_set, test_label);

[er2, bad] = nntest(nn, train_set, train_label);

disp(['Classification accuracy on training set: ',num2str((1-er1)*100),'%']);

disp(['Classification accuracy on test set: ',num2str((1-er2)*100),'%']);

Thank you in advance!

[Rohit Kulkarni]

is nntrain a matlab function?

[CHRISTINA]

Yes, it’s a deeplearn toolbox function.

Answer 1

[Yatharth]

Hey I am not sure about your approch of creating a Neural Network, you can follow up this approch mentioned in the documentation for the MNIST Data Set . Alternatively if you want to build your own neural network you can follow this tutorial