Analysis (evaluation) is very slow on large datasets

Just realised that the evaluation procedure is rather slow for a large dataset such as CXR8. We need to try and improve that with multiprocessing, possibly.