Skip to content

README.md

Latest commit

 

History

History
63 lines (41 loc) · 4 KB

README.md

File metadata and controls

63 lines (41 loc) · 4 KB
Raw

ZunZunNG

A Django site for 2D and 3D nonlinear curve and surface fitting via the pyeq3 library — with genetic-algorithm initial parameter estimation, orthogonal-distance and relative-error regressions, source-code generation in several languages, PDF reports, and surface animations.

Hosted at github.com/kiloscheffer/zunzun-ng.

What it does

  • Fits a large catalogue of named 2D and 3D equations to user-supplied data.
  • Uses Differential Evolution (a genetic algorithm) to find good starting points before handing off to a non-linear solver — so you don't need to guess initial parameter values.
  • Ranks candidate equations by AIC, SSQ, and other fit statistics (the Function Finder flow).
  • Emits fitted-coefficient source code in C++, C#, Fortran 90, Java, JavaScript, Julia, and Python.
  • Generates a PDF report per fit (cover page, statistics, scatterplots, histograms) and animated GIFs for 3D surface visualisations.
  • Handles outliers, weighted fits, splines, and user-defined functions.

Development setup

Dependencies are managed with uv. To install uv, see the uv install guide — or on Debian/Ubuntu: apt-get install pipx && pipx install uv.

uv sync                                   # create .venv, install deps
uv run python manage.py migrate           # one-time: create the session DB
uv run python manage.py runserver         # http://127.0.0.1:8000/

Production

uv sync --no-dev
uv run waitress-serve --listen=127.0.0.1:8000 wsgi:application

Per-platform recipes (systemd unit, launchd plist, IIS + NSSM) live in docs/deployment/. Waitress is the recommended cross-platform WSGI server; legacy gunicorn / Apache + mod_wsgi setups still work on Linux but require threads=1 to avoid fork-safety hazards in view code that spawns subprocesses.

Testing

uv run pytest tests/                      # 295 unit tests, ~25s
uv run python scripts/smoke_test.py       # end-to-end smoke, ~1–5 min

The smoke script starts a throwaway Waitress, POSTs a real fit, polls for completion, and asserts structural markers in the result. The legacy FunkLoad suite under funkload_tests/ is not runnable under the uv-managed Python 3.14 environment.

Cross-platform

ZunZunNG runs natively on Linux, macOS, and Windows. The original os.fork() architecture was replaced with multiprocessing.Process(spawn) so the code no longer depends on Unix-style process forking. Platform-specific calls (load average, process priority, zombie reap) are isolated in zunzun/platform_compat.py.

About this fork

ZunZunNG ("Next Generation") is a permanent fork of James R. Phillips's zunzunsite3 (Copyright © 2016 James R. Phillips, BSD-2-clause; dormant since 2020), modernized for:

  • Python 3.14 / Django 6.0 (was Python 3.x / Django 2.2).
  • Cross-platform deployment (Linux, macOS, Windows) via the spawn-based multiprocessing migration.
  • scipy.odr removal — replaced with the independent odrpack package (scipy.odr was deprecated in scipy 1.17, removed in 1.19). Done first in the companion pyeq3-ng fork and since merged upstream (PR #8); ZunZunNG now depends on upstream pyeq3 directly.
  • Pure-Python animated GIFs via matplotlib's PillowWriter, replacing the previous imagemagick + gifsicle system-binary dependency.

The original copyright notice in LICENSE.txt is retained per the BSD-2-clause terms. James R. Phillips's prose in the "About" page is preserved verbatim.

For deeper notes on the spawn-based long-running-process pattern, the parallel session stores, and the ChildPayload contract, see docs/internals/active-gotchas.md.

License

BSD-2-clause. See LICENSE.txt.