Randomness is now generated via a numpy.random Generator. Anything that relied on using numpy.random.seed to obtain deterministic results will fail (mostly of relevance for testing). Seeds (or Generators) can now be passed to relevant functions with the seed argument, but note that changes in Numpy's random number generation means this will not give the same results that would be obtained using numpy.random.seed to set the seed in previous versions. There is a new pval function, and there are several bugfixes. Numba is now supported in some instances (np.average or np.mean as statfunction, 1-D data), using use_numba=True. Pypy3 is also supported. Typing information has been added, with code passing mypy --strict --allow-untyped-calls --ignore-missing-imports, and tests cover 100% of the code (though many tests use fixed seeds). Handling of multiple data sets (tuples/etc of arrays) now can be specified as multi="paired" (the previous handling), where the sets must be of the same length, and samples are taken keeping corresponding points connected, or multi="independent", treating data sets as independent and sampling them seperately (in which case they may be different sizes).