multidimensionalks 0.0.7

Multidimensional KS test module in python

multidimensionalks

Python c extension with method for calculating multidimensional Kolmogorov-Smirnov test

multidimensionalks.test(rvs, cdf=None, counts_rvs=None, counts_cdf=None, n_jobs=1, permutation_samples=0, binomial_significance=False, use_avx=True, max_alpha_beta=False, scale_result=True, deduplicate_data=True)

Example usage

from multidimensionalks import test
import numpy as np

test(np.array([[1, 2, 3], [1, 3, 2]]), cdf=np.array([[1,2,2]]))

Parameters

• rvs: 2-dimensional numpy number array with rows representing d-dimensional observations,
• cdf: 2-dimensional numpy number array with rows representing second sampled-dimensional observations,
• counts_rvs: in case of rvs having multiple duplicates, an array without duplicates and a separate array of counts can be provided,
• counts_cdf: in case of cdf having multiple duplicates, an array without duplicates and a separate array of counts can be provided, additionally if cdf is not given counts_cdf are taken as counts of elements of rvs array,
• n_jobs: number of threads used during calculation,
• permutation_samples: number of times data is shuffled and the statistic value is calculated to estimate pvalue,
• binomial_significance: boolean value indicating if statistical significance should be calculated. Defaults to False,
• use_avx: boolean value indicating if AVX2 instructions should be used during the calculations. Defaults to True,
• max_alpha_beta: boolean value indicating how λ and β values should be combined. Value True results in max(λ, β). λ+β is used otherwise.
• scale_result: Whether to scale the statistic by $\frac{1}{2}\sqrt{\frac{|rvs|+|cdf|}{|rvs|\times|cdf|}}$,
• deduplicate_data: Whether to deduplicate data points before running the algorithms.

Return value

If no pvalue calculation method is selected returns ks statistic value, otherwise returns a tuple:

• ks statistic,
• pvalue calculated using statistical method if binomial_significance is set to True,
• pvalue calculated using permutation method if permutation_samples is larger than 0.