Open axch opened 9 years ago
Relevant code locations:
Much though I like the algorithmic challenge of doing a K-S test for agreement between a sampler and a probability density function by running the quadrature needed to compute the cdf from the pdf in tandem with the cdf queries the K-S test requires and cleverly reusing query points, probably the cleanest way to achieve the desired effect for 1-D continuous distributions would be to just reuse analytically known cdfs.
To wit:
Finite-domain discrete distributions are easier, because the exact answer for the Chi^2- or G-test can be computed by enumerating all values, because the pdf is also the probability mass function.