EN • ES

Fit probability distributions online with Phitter

BETA

Phitter analyzes datasets and determines the best analytical probability distributions that represent them. The Phitter kernel studies over 80 probability distributions, both continuous and discrete, 3 goodness-of-fit tests, and interactive visualizations. For each selected probability distribution, a standard modeling guide is provided along with spreadsheets that detail the methodology for using the chosen distribution in data science, operations research, and artificial intelligence.

Phitter is exhaustive and free

Get StartedDistributions

Preparing the demonstration…

Free, open-source statistical software

Find a probability distribution for your data.

Phitter is free distribution fitting software that compares probability distributions with your observations. Explore the estimated parameters, goodness-of-fit tests and plots in one place, then choose a model that makes sense for your data.

Use the online distribution fitting tool without installation or an account. For repeatable analyses in a notebook, script or simulation workflow, use the Phitter Python library.

Fit a dataset Read the guide

From observations to a fitted model

  1. Bring your sample

    Paste values or import a TXT, CSV or XLSX file. Choose continuous or discrete data and review the sample before fitting.

    At least 20 observations. Larger samples may help distinguish plausible models.

  2. Compare the candidates

    Phitter estimates parameters and runs the applicable goodness-of-fit tests. Continuous fits use BIC, AIC and SSE to break ties after the number of tests not rejected; discrete fits use SSE.

    Each result includes the fitted parameters and test outcomes.

  3. Choose a defensible model

    Inspect the plots and compare the leading candidates. Consider the range of possible values, tail behaviour and the process that generated your observations.

    The first row is a candidate for investigation, not proof of the data-generating distribution.

Read the evidence behind the ranking.

A single score cannot answer every modelling question. Use the numerical results and visual checks together.

Estimated parameters
Inspect the location, scale and shape of each fitted model, where applicable.
Goodness-of-fit tests
Compare Kolmogorov–Smirnov, Anderson–Darling and chi-square results where applicable. A test that does not reject a model does not establish that it is correct.
Visual diagnostics
Compare your observations with the fitted density or mass function, cumulative distribution and Q–Q plot.
How to read a ranking

Fitting runs in your browser.

The Python kernel runs locally through Pyodide. Your observations are not uploaded to a Phitter fitting server, and you do not need an account. The site also uses analytics and external services, described in the privacy policy.

Privacy policy

Continue in Python.

Use the open-source library when you need to repeat an analysis, control candidate distributions or include fitting in a larger workflow. Record your version, data preparation and fitting options so the result can be reproduced.

Python documentation

Methods you can inspect

Created by Sebastián José Herrera Monterrosa. Phitter has a peer-reviewed software paper in JOSS, with source code and documentation available to explore.

About the project