Project Description
## Example

## Simple tests

## Extending

## Installation

Release History
## Release History

Donec et mollis dolor. Praesent et diam eget libero egestas mattis sit amet vitae augue. Nam tincidunt congue enim, ut porta lorem lacinia consectetur. Donec ut libero sed arcu vehicula ultricies a non tortor. Lorem ipsum dolor sit amet, consectetur adipiscing elit.

Download Files
## Download Files

Provides variants of Kolmogorov-Smirnov, Cramer-von Mises and Anderson-Darling goodness of fit tests for fully specified continuous distributions.

>>> from scipy.stats import norm, uniform >>> from skgof import ks_test, cvm_test, ad_test >>> ks_test((1, 2, 3), uniform(0, 4)) GofResult(statistic=0.25, pvalue=0.97...) >>> cvm_test((1, 2, 3), uniform(0, 4)) GofResult(statistic=0.04..., pvalue=0.95...) >>> data = norm(0, 1).rvs(random_state=1, size=100) >>> ad_test(data, norm(0, 1)) GofResult(statistic=0.75..., pvalue=0.51...) >>> ad_test(data, norm(.3, 1)) GofResult(statistic=3.52..., pvalue=0.01...)

Scikit-gof currently only offers three nonparametric tests that let you compare a sample with a reference probability distribution. These are:

`ks_test()`- Kolmogorov-Smirnov supremum statistic; almost the same as
`scipy.stats.kstest()`with`alternative='two-sided'`but with (hopefully) somewhat more precise p-value calculation; `cvm_test()`- Cramer-von Mises L2 statistic, with a rather crude estimation of the statistic distribution (but seemingly the best available);
`ad_test()`- Anderson-Darling statistic with a fair approximation of its distribution;
unlike the composite
`scipy.stats.anderson()`this one needs a fully specified hypothesized distribution.

Simple test functions use a common interface, taking as the first argument the
data (sample) to be compared and as the second argument a frozen `scipy.stats`
distribution.
They return a named tuple with two fields: `statistic` and `pvalue`.

For a simple example consider the hypothesis that the sample (.4, .1, .7) comes from the uniform distribution on [0, 1]:

if ks_test((.4, .1, .7), unif(0, 1)).pvalue < .05: print("Hypothesis rejected with 5% significance.")

If your samples are very large and you have them sorted ahead of time, pass
`assume_sorted=True` to save some time that would be wasted resorting.

Simple tests are composed of two phases: calculating the test statistic and determining how likely is the resulting value (under the hypothesis). New tests may be defined by providing a new statistic calculation routine or an alternative distribution for a statistic.

Functions calculating statistics are given evaluations of the reference cumulative distribution function on sorted data and are expected to return a single number. For a simple test, if the sample indeed comes from the hypothesized (continuous) distribution, the values passed to the function should be uniformly distributed over [0, 1].

Here is a simplistic example of how a statistic function might look like:

def ex_stat(data): return abs(data.sum() - data.size / 2)

Statistic functions for the provided tests, `ks_stat()`, `cvm_stat()`,
and `ad_stat()`, can be imported from `skgof.ecdfgof`.

Statistic distributions should derive from `rv_continuous` and implement
at least one of the abstract `_cdf()` or `_pdf()` methods (you might
also consider directly coding `_sf()` for increased precision of results
close to 1). For example:

from numpy import sqrt from scipy.stats import norm, rv_continuous class ex_unif_gen(rv_continuous): def _cdf(self, statistic, samples): return 1 - 2 * norm.cdf(-statistic, scale=sqrt(samples / 12)) ex_unif = ex_unif_gen(a=0, name='ex-unif', shapes='samples')

The provided distributions live in separate modules, respectively `ksdist`,
`cvmdist`, and `addist`.

Once you have a statistic calculation function and a statistic distribution the
two parts can be combined using `simple_test`:

from functools import partial from skgof.ecdfgof import simple_test ex_test = partial(simple_test, stat=ex_stat, pdist=ex_unif)

**Exercise**: The example test has a fundamental flaw. Can you point it out?

pip install scikit-gof

Requires recent versions of Python (> 3), NumPy (>= 1.10) and SciPy.

Please fix or point out any errors, inaccuracies or typos you notice.

TODO: Figure out how to actually get changelog content.

Changelog content for this version goes here.

Donec et mollis dolor. Praesent et diam eget libero egestas mattis sit amet vitae augue. Nam tincidunt congue enim, ut porta lorem lacinia consectetur. Donec ut libero sed arcu vehicula ultricies a non tortor. Lorem ipsum dolor sit amet, consectetur adipiscing elit.

TODO: Figure out how to actually get changelog content.

Changelog content for this version goes here.

Donec et mollis dolor. Praesent et diam eget libero egestas mattis sit amet vitae augue. Nam tincidunt congue enim, ut porta lorem lacinia consectetur. Donec ut libero sed arcu vehicula ultricies a non tortor. Lorem ipsum dolor sit amet, consectetur adipiscing elit.

TODO: Figure out how to actually get changelog content.

Changelog content for this version goes here.

Donec et mollis dolor. Praesent et diam eget libero egestas mattis sit amet vitae augue. Nam tincidunt congue enim, ut porta lorem lacinia consectetur. Donec ut libero sed arcu vehicula ultricies a non tortor. Lorem ipsum dolor sit amet, consectetur adipiscing elit.

TODO: Figure out how to actually get changelog content.

Changelog content for this version goes here.

TODO: Brief introduction on what you do with files - including link to relevant help section.

File Name & Checksum SHA256 Checksum Help | Version | File Type | Upload Date |
---|---|---|---|

scikit-gof-0.1.2.tar.gz (10.2 kB) Copy SHA256 Checksum SHA256 | – | Source | Apr 10, 2016 |