Quickstart¶

Fit a single Gaussian peak to synthetic data in a few lines. This example uses only the public spectrafit_core API: the gaussian() node factory, compose() to build a FitGraph, MeasurementData for the input, and fit() to run the solver.

import numpy as np
from spectrafit_core import MeasurementData, compose, fit, gaussian

# Synthetic "measured" data: a Gaussian peak plus a little noise.
x = np.linspace(-5, 5, 200)
y = 3.0 * np.exp(-0.5 * ((x - 0.5) / 1.2) ** 2) + np.random.default_rng(0).normal(
    0, 0.05, x.size
)

# Build the model graph: one Gaussian node with initial guesses.
graph = compose([gaussian("peak1", amplitude=1.0, center=0.0, sigma=1.0)]).build()

# Run the fit.
result = fit(graph, MeasurementData(x=x.tolist(), y=y.tolist()))

print(result.parameters)  # fitted amplitude/center/sigma, keyed "peak1.<param>"
print(result.r_squared)  # goodness of fit

Expected output¶

The noise is drawn from a seeded generator (default_rng(0)), so this example is deterministic — you should see these numbers exactly:

{'peak1.sigma': ParameterResult(value=1.2024163241228947, min=-inf, max=inf, vary=True,
                                expr=None, scale=None, name='peak1.sigma',
                                stderr=0.004211511681961483),
 'peak1.amplitude': ParameterResult(value=2.997594452290324, ...,
                                stderr=0.00909249987508253),
 'peak1.center': ParameterResult(value=0.49756589193680784, ...,
                                stderr=0.004211469158787863)}
0.9979075074894731

The block above is reformatted for readability: print emits it as one long line, and dict ordering is an implementation detail — your keys may appear in a different order. The values are what matter, and they are reproducible: the example seeds its RNG.

The fit is worth reading against the data it was given. The synthetic peak was built with amplitude 3.0, centre 0.5 and sigma 1.2, and the solver recovers 2.9976, 0.4976 and 1.2024 from initial guesses of 1.0, 0.0 and 1.0 — each within about one standard error of the truth, in 7 iterations. The remaining difference is the noise, not the solver: r² = 0.9979 against data carrying 0.05 of added noise on a peak of height 3.

Every parameter comes back as a ParameterResult rather than a bare float, so the bounds (min/max), whether it was fitted or held (vary), any tie (expr), and its standard error travel with the value.

fit() returns a FitResult with the fitted parameters, uncertainties, and goodness-of-fit statistics (r_squared, chi2, reduced_chi2, aic, bic, …). For a variant that also returns the best-fit curve directly as a NumPy array (skipping JSON round-tripping of per-point arrays), use fit_fast() instead of fit().

Next steps¶

This example is deliberately minimal — one peak, default solver, no bounds or ties. For multi-peak fits, bounds, shared/tied parameters, alternative solvers, and 2D/N-D data, see the tutorial gallery.