Please see pull request: #2

awarp sometimes returns np.nan for calculations since values in the distance matrix [-1,-1] are negative and the square root fails. This is related to an uncaught overflow in numpy or numba

Thanks for the implementation. I'd be interested to use it in a project on sparse binary sequences. However, I'm not yet sure about your expected format. You provide these two arrays here as examples:

s = np.array([1, 2, 3, -1, 1])
t = np.array([1, -2, 4, 1])

However, I don't get how these match either the run-length encoding of the paper (then I'd expect numbers highlighted somehow to give an idea of the number of zeros) nor the encoding described as in Wikipedia, as then I'd expect both to be of even length, as it would be alternating between the count and the value (i.e., decoded t would be t = np.array([-2, 1,1,1,1]), but s would not be clear due to uneven length).

Can you provide an example of the original time series and the encoded ones?