makedatabatches.m

function [Xtrainb, ytrainb, Xtestb, ytestb] = makedatabatches(X, y, train_percent)
%MAKEDATABATCHES shuffle data, partition the data into training and testing datasets,
% then partition each set into batches for dbn training

N = size(X,1);
shuffle = randperm(N);
Ntrain = ceil(train_percent * N);
Ntest = N - Ntrain;
Xtrain = X(shuffle(1:Ntrain),:);
ytrain = y(shuffle(1:Ntrain),:);
Xtest = X(shuffle(Ntrain+1:end),:);
ytest = y(shuffle(Ntrain+1:end),:);

batch_size = 1000;
[train_batch_size, last_train_batch] = deal(floor(Ntrain/batch_size), mod(Ntrain, batch_size));
[test_batch_size, last_test_batch] = deal(floor(Ntest/batch_size), mod(Ntest, batch_size));
Xtrainb = mat2cell(Xtrain, [batch_size*ones(1,train_batch_size), last_train_batch]);
ytrainb = mat2cell(ytrain, [batch_size*ones(1,train_batch_size), last_train_batch]);
Xtestb = mat2cell(Xtest, [batch_size*ones(1,test_batch_size), last_test_batch]);
ytestb = mat2cell(ytest, [batch_size*ones(1,test_batch_size), last_test_batch]);

%{
Earlier implementation using truncated matrices
[num_train_batches, train_batch_size] = deal(floor(sqrt(Ntrain)));
[num_test_batches, test_batch_size] = deal(floor(sqrt(Ntest)));
assert(num_train_batches > 0 && num_test_batches > 0, 'Not enough training/testing data for one batch');
Xtrainb = permute(reshape(Xtrain(1:num_train_batches^2,:)', [F, num_train_batches, train_batch_size]), [2,1,3]);
%}

end