* Consider calculating MI between *relative* axial positions or velocities of pairs of particles over time,
rather than branching histories, which also might work in the infinite-time limit.
-* Parallelise in something natural, such as redis or 0MQ, or ruffus. Probably ruffus + sqlite is easiest.
+* Parallelise in something natural, such as redis or 0MQ, or ruffus.
+* store data in sqlite for ease of processing
+* store intermediate trials in sqlite too
+* calibrate pyentropy to work out why i get large positive information between deterministic variables.
from __future__ import division
if n_samples<20 and test:
raise ValueError("%d is too small a number to bin" % n_samples)