Result accuracy

Although the algorithm returns exact results in theory, in practice it returns result only exact up to a given precision. The epsilon keyword argument allows to control that precision, it is set to 1e-7 by default.

>>> import numpy
>>> import miniball
>>> S = numpy.random.randn(100, 2)
>>> C, r2 = miniball.get_bounding_ball(S, epsilon=1e-7)

Repeatability

The algorithm to compute bounding balls relies on a pseudo-random number generator. Although the algorithms return an exact solution, it is only exact up to the epsilon parameter. As a consequence, running the get_bounding_ball function twice on the same input might not return exactly the same output.

By default, each call to get_bounding_ball pull out a new, freshly seeded pseudo-random number generator. Therefore, if you wish to get repeatable results from get_bounding_ball, you have to (and only have to) pass the same pseudo-random number generator, using with the rng keyword argument

>>> import numpy
>>> import miniball
>>> S = numpy.random.randn(100, 2)
>>> rng = numpy.random.RandomState(42)
>>> C, r2 = miniball.get_bounding_ball(S, rng = rng)

Duplicated input points

Duplicated input points might trigger failures. This implementation do not check for duplicated input points, the guaranty of non-duplication is defered to you, the programmer. This is by design, to avoid to pay the cost of de-duplication when we are sure that the input has no duplicates.