Supsmu

Supsmu is an implementation of Friedman's SuperSmoother algorithm - a time series smoother that uses cross-validation to automatically select optimal spans for local linear regression.

The package is written in C for computational efficiency, with Python bindings for use with NumPy arrays.

Installation

Install supsmu with:

pip install supsmu

Python Example

Minimal example using dummy data:

import numpy as np
from supsmu import supsmu

x = np.linspace(0, 1, 100)
# Dummy data - a basic sine wave
y = np.sin(2 * 2 * np.pi * x)
noise = np.random.normal(0, 0.2, 100)
y_noisy = y + noise

y_smooth = supsmu(x, y_noisy, periodic=True)

Additional Information

The C algorithm is implemented with the intent to closely match the outputs of the Fortran version (available in R), however there may still exist some inconsistencies.

Parameters

PARAMETER	TYPE	DESCRIPTION
x	np.ndarray[np.floating | np.integer]	x values
y	np.ndarray[np.floating | np.integer]	y values
wt	np.ndarray[np.floating | np.integer] | None	weights
span	float	smoothing span (0 for cross-validation, otherwise between 0 and 1)
periodic	bool	True if data is periodic, False otherwise
bass	float	bass enhancement (between 0 and 10) for increased smoothness

References

[1] J. H. Friedman, "A Variable Span Smoother", SLAC National Accelerator Laboratory (SLAC), Menlo Park, CA (United States), SLAC-PUB-3477; STAN-LCS-005, Oct. 1984. doi: 10.2172/1447470.