distfitlab

Interactive distribution fitting in Jupyter. Simulate or load data, fit many probability distributions at once, see which one fits best, and inspect the data — from a single widget UI.

Two apps behind one switch:

• Continuous — 18 distributions: Normal, Exponential, Gamma, Rayleigh, Weibull, Lognormal, Loglogistic, Inverse Gaussian, Beta, GEV, Logistic, Laplace, Chi-squared, Chi, Nakagami, Rician, Cauchy, Student-T. Fitted with scipy.stats.
• Discrete — 5 count distributions: Poisson, Binomial, Geometric, Negative Binomial, Zero-Inflated Poisson. SciPy has no .fit for these, so the fitters are written here.

Install

pip install distfitlab

Optional readers and the interactive run charts are extras:

pip install "distfitlab[all]"

For the latest development version, install from GitHub instead: pip install "distfitlab[all] @ git+https://github.com/rberangi/distfitlab.git".

After installing

distfitlab is a widget app, so it runs inside a notebook — not in a plain Python script or terminal.

1. Open a notebook in the same Python environment you installed into:

pip install jupyterlab      # skip if you already have Jupyter
jupyter lab

VS Code works too: create a .ipynb file and select that Python as the kernel. Notebook 7 is also fine — anywhere ipywidgets renders.

2. Start the app in a notebook cell:

from distfitlab import main

main()                # opens on the continuous app
# main("Discrete")    # opens on the discrete app

3. Use the UI that appears below the cell:

1. Data source — simulate data, or choose From file to upload a file (or enter a path) and pick the column.
2. Fit All — fit every distribution and get a table ranked best-first.
3. Fit — try one distribution with parameters you type.
4. Visualize data — histogram, ECDF, box plot, Q-Q plot, run charts and summary stats.
5. Save results — write the table and figures to an outputs/ folder next to the notebook.

A ready notebook is in examples/quickstart.ipynb (in the repository; a pip install does not include it), and every control is explained in docs/guide.md.

If something doesn't work

What you get

• Fit All fits every distribution and ranks them best-first by the largest gap between the empirical and theoretical CDF, with the parameters and AIC/BIC for each.
• Fit scores one distribution against parameters you type, with CDF/PDF overlays.
• Data view — histogram, ECDF, empirical PDF (counts: empirical PMF), box plot, Q-Q plot against the distribution you chose, run chart and sorted run chart, and summary statistics. ECDF, PDF/PMF and the run charts have a pan/zoom toolbar and optional grid lines.
• File data — CSV, TSV, Excel, JSON, Parquet, Feather, SPSS/Stata/SAS and HDF5; a stack of filter conditions across columns; cleaning (drop NaN, drop ≤ 0, percentile trim, drop duplicates); and Group by, which fits every group and ranks them.
• loc = 0 option, for when a distribution should be anchored at the origin.
• Conditional probability — P(event | filters) for an event on the Use column, e.g. P(count > 500 | season == summer), with its all-rows baseline, the ratio, Bayes' reverse P(condition | event), and the model's value.
• Save results writes the table, the samples and the figures to a timestamped folder.

Requirements

Python 3.9+, with numpy, pandas, scipy, matplotlib and ipywidgets. Developed and tested on Python 3.12.

Documentation

The full walkthrough — every control, the parameter conventions, supported file types, troubleshooting and how the fitting works — is in docs/guide.md.

License

MIT — see LICENSE.

Citation

If this helps with published work, please cite the libraries it stands on: SciPy, NumPy, pandas, Matplotlib and ipywidgets.