distribution_discovery.py

#!/usr/bin/env python

from __future__ import print_function
import numpy as np


def grad_norm(array, norm='l1'):
    grad = np.gradient(array)

    if norm == 'l1':
        integrand = np.absolute(grad)
    elif norm == 'l_infty':
        integrand = np.max(np.absolute(grad))
    else:
        print('Using l2 norm')
        integrand = np.square(grad)

    return np.sum( integrand )


def choose_distribution(*dists, **kwargs):
    """! Chooses a distribution for a model based on the test outlined in Dewhurst et al. 2018

    @param dists (list): list of distributions to check. These are intended to be generated by 
    the diffusion transform action on a the particular model.
    
    Optional parameters

    @kwarg norm (str): the type of norm to use. Should be one of 'l1', 'l2', or 'l_infty'.
    """
    
    norm = kwargs.get('norm', 'l1')
    vals = {}

    for dist in dists:
        test_val = grad_norm(dist, norm=norm)
        vals[f.__name__] = test_val

    return vals