Below is an example of how to use "bootstrp" from the "Statistics and Machine Learning Toolbox" to perform non-overlapping block resampling on a MATLAB array:
ts = ["A","B","C","D"]
blockSize = 2;
numBlocks = length(ts) / blockSize; % must be integer
blocks = reshape(ts, [numBlocks,blockSize])' % reshape into non-overlapping blocks
nSamples = 10;
samples = bootstrp(nSamples, @(x)x', blocks) % @(x)x' maintains the sequence of items in the block
Running the code above will demonstrate that the "samples" output contains randomly sampled permutations of the rows in "blocks". In this case the blocks are "AB" and "CD", for ease of demonstration.
The main limitation of this approach is that the number of data points in the time series must be evenly divisible by the block size or "reshape" will throw an error.
