distributions.py

# This code is heavily based on a blog post by Andre Dietrich, currently
# found at http://www.aizac.info/simple-check-of-a-sample-against-80-distributions/

from __future__ import print_function

import scipy.stats
import warnings
import numpy as np
import sys
import operator

# just for surpressing warnings
warnings.simplefilter('ignore')

t = sys.argv[1:]
stats = {}
for filename in t:
    print("processing %s" % filename)
    data = np.loadtxt(filename)
    cdfs = [func for func in dir(scipy.stats) if callable(getattr(scipy.stats, func))]
    for cdf_name in cdfs: 
        try:
            cdf = eval("scipy.stats."+cdf_name)
            #fit our data set against every probability distribution
            parameters = cdf.fit(data)
            negative_log_likelihood = cdf.nnlf(parameters, data)
            aicc = 2 * data.size * len(parameters) / (data.size - len(parameters) - 1) + 2 * negative_log_likelihood
            if not np.isnan(aicc) and not np.isinf(aicc):
                stats[cdf_name] = aicc
        except (AttributeError, NotImplementedError, TypeError):
            pass
            
# Sort
sorted_stats = sorted(stats.items(), key=operator.itemgetter(1))


for cdf, aicc in sorted_stats:
    print("{:15s} {:10.2f}".format(cdf, aicc))