fitwise

One call. One fit. One report you can trust.

Status: alpha (0.2). Fitting, built-in and text models, automatic starting values, robust fits, report, plot and export work and are tested against the NIST reference data. The API may still change before 1.0. Documentation: https://maroba.github.io/fitwise/

fitwise is an ergonomic layer over lmfit and scipy for curve fitting in the lab. It brings no new solver of its own. Instead it makes the existing ones fast to use and hard to misuse:

import fitwise as fw

result = fw.fit(fw.models.gaussian, x, y, yerr=dy)
print(result)              # report: parameters ± errors, χ²/ndf, R², correlations, warnings
result.plot()              # data, fit, confidence band, residuals
result.params["sigma"]     # value, uncertainty, unit

What it does for you

• Picks the solver: direct linear least squares when the model is linear in its parameters, Levenberg–Marquardt otherwise, and orthogonal distance regression when xerr is given.
• Finds starting values, including for your own functions and for models written as text (fw.fit("a*exp(-x/tau) + c", x, y)), and shows how it found them.
• Resists outliers on request: loss="huber" downweights them and marks them in the plot.
• Gives honest uncertainties: a documented covariance convention, confidence and prediction bands, and bootstrap or profile likelihood on request.
• Warns instead of returning wrong numbers: non-convergence, parameters at a bound, poor χ²/ndf and strong correlations all show up in the report.
• Makes fits reproducible: result.to_dict() captures everything needed to repeat the fit.

Installation

pip install fitwise            # core: numpy, scipy, lmfit
pip install "fitwise[all]"     # + plotting, ODR, units, symbolic models, batch fits

Development

uv sync --all-extras           # create .venv with all extras and dev tools
uv run pre-commit install
uv run pytest                  # fast tests
uv run pytest -m nist          # NIST StRD reference suite

License

BSD-3-Clause