Providing cross-validation sets for LSTM

[shaimaahegazy]

Is there a way to provide cross-validation/test sets for the LSTM training function? I can understand that the function is defined as follows:

gradientDescentAdaDelta(costFunc, nn, defs, Xts, Yts, yts, y, 'Training
Entire Network');

My question is what are the : Xts, Yts, yts and y variables?

Thanks a lot.

[guanyou]

From what I know, like for the MNIST example data, X being the training data and y being the testing data,

X = (28_28... image data)= 784 * 60000 .... variations training data input
Xts = (28_28...image data)= 784 * 10000... variations test data input
y =(label of image data) 1 * 60000...variations training data output
yts = (label for image data)1 * 10000 .. .variations test data output

[shaimaahegazy]

Thanks for your reply @guanyou. Do you know where I could get the MNIST data used in the example?

[guanyou]

I'm guessing u could get it from below, or just implement those data conversion code found online with the original dataset found here http://yann.lecun.com/exdb/mnist/

http://www.cs.toronto.edu/~norouzi/research/mlh/data/mnist-full.mat
http://cs.nyu.edu/~roweis/data.html

[RMano82]

Hi @shaimaahegazy . I wonder if you have managed to run any of the examples provided with the toolbox where the network converges.

I, so far tried the MNIST_ConvNet_Classifier.m and the MNIST_Deep_Classifier.m where I download the mnist-full.mat from
http://www.cs.toronto.edu/~norouzi/research/mlh/data/mnist-full.mat
However, the networks in both examples do not converge.
Any help will be much appreciated