galvanized / fitter

Fit data to many distributions

FITTER documentation

Compatible with Python 3.6, 3.7, and 3.8, 3.9

What is it ?

fitter package provides a simple class to identify the distribution from which a data samples is generated from. It uses 80 distributions from Scipy and allows you to plot the results to check what is the most probable distribution and the best parameters.

Installation

pip install fitter

fitter is also available on conda (bioconda channel):

conda install fitter

And as a singularity file in damona <https://damona.readthedocs.io>:

pip install damona
damona install fitter

Usage

standalone

A standalone application (very simple) is also provided and works with input CSV files:

fitter fitdist data.csv --column-number 1 --distributions gamma,normal

It creates a file called fitter.png and a log fitter.log

From Python shell

First, let us create a data samples with N = 10,000 points from a gamma distribution:

from scipy import stats
data = stats.gamma.rvs(2, loc=1.5, scale=2, size=10000)

Note

the fitting is slow so keep the size value to reasonable value.

Now, without any knowledge about the distribution or its parameter, what is the distribution that fits the data best ? Scipy has 80 distributions and the Fitter class will scan all of them, call the fit function for you, ignoring those that fail or run forever and finally give you a summary of the best distributions in the sense of sum of the square errors. The best is to give an example:

from fitter import Fitter
f = Fitter(data)
f.fit()
# may take some time since by default, all distributions are tried
# but you call manually provide a smaller set of distributions
f.summary()

See the online documentation for details.

Changelog

Version	Description
1.4.0	get_best function now returns the parameters as a dictionary of parameter names and their values rather than just a list of values (cokelaer#23) thanks to contributor @kabirmdasraful Accepting PR to fix progress bar issue reported in cokelaer#37
1.3.0	parallel process implemented cokelaer#25 thanks to @arsenyinfo
1.2.3	remove vervose arguments in Fitter class. Using the logging module instead the Fitter.fit has now a progress bar add a standalone application called … fitter (see the doc)
1.2.2	was not released
1.2.1	adding new class called histfit (see documentation)
1.2	Fixed the version. Previous version switched from 1.0.9 to 1.1.11. To start a fresh version, we increase to 1.2.0 Merged pull request required by bioconda Merged pull request related to implementation of AIC/BIC/KL criteria (cokelaer#19). This also fixes cokelaer#9 Implement two functions to get all distributions, or a list of common distributions to help users decreading computational time (cokelaer#20). Also added a FAQS section. travis tested Python 3.6 and 3.7 (not 3.5 anymore)
1.1	Fixed deprecated warning fitter is now in readthedocs at fitter.readthedocs.io
1.0.9	cokelaer#8 and 11 PR cokelaer#8
1.0.6	summary() now returns the dataframe (instead of printing it)
1.0.5	https://github.com/cokelaer/fitter/issues
1.0.2	add manifest to fix missing source in the pypi repository.