chronax.loess

LOESS/LOWESS (Locally Weighted Scatterplot Smoothing) in JAX. This module implements LOESS (Local Polynomial Regression) with tricube kernel weighting and optional robust reweighting for non-parametric smoothing of time series and scatter data.

loess_window_jump

chronax.loess.loess_window_jump

LOESS with fixed odd window using tricube weights. - deg ∈ {0,1}: local constant or local linear (value at center). - robust_outer: bisquare outer reweighting count. - jump: compute only at anchor points and linearly interpolate in-between (STL-like speedup).

loess_window_jump(y, window, deg=1, robust_outer=0, jump=1)

Parameter	Type	Default	Description
y	jnp.ndarray	-	Input 1D array to smooth.
window	int	-	Window size (will be forced to odd).
deg	int	1	Polynomial degree, 0 (local constant) or 1 (local linear).
robust_outer	int	0	Number of robust reweighting iterations using bisquare weights.
jump	int	1	Evaluate only at anchor points (0, jump, 2*jump, ...) and interpolate.

Returns: jnp.ndarray (Smoothed values of shape (n,).)

lowess_frac

chronax.loess.lowess_frac

Statsmodels-style LOWESS (Cleveland): Neighborhood defined by frac*n nearest x. Tricube kernel; robust outer iterations via bisquare. Returns yhat aligned to x (input order).

lowess_frac(y, x, frac=0.6666666666666666, it=0)

Parameter	Type	Default	Description
y	jnp.ndarray	-	Input 1D array of values to smooth.
x	jnp.ndarray	-	Input 1D array of x-coordinates (arbitrary order).
frac	float	2.0 / 3.0	Fraction of data used for neighborhood (0 < frac <= 1).
it	int	0	Number of robust reweighting iterations.

Returns: jnp.ndarray (Smoothed values of shape (n,) aligned to input order.)

loess_smooth

chronax.loess.loess_smooth

Convenience wrapper. Choose one mode: Fixed-window: pass window (odd). Uses loess_window_jump (supports jump, deg, robust_outer). Frac-based: pass frac and x. Uses lowess_frac (supports robust iterations).

loess_smooth(y, *, window=None, frac=None, deg=1, robust_outer=0, jump=1, x=None)

Parameter	Type	Default	Description
y	jnp.ndarray	-	Input 1D array to smooth.
window	int \| None	None	Window size for fixed-window LOESS.
frac	float \| None	None	Fraction of data for neighborhood in LOWESS mode.
deg	int	1	Polynomial degree (0 or 1) for fixed-window mode.
robust_outer	int	0	Number of robust reweighting iterations.
jump	int	1	Jump size for anchor points in fixed-window mode.
x	jnp.ndarray \| None	None	X-coordinates for LOWESS mode.

Returns: jnp.ndarray (Smoothed values of shape (n,).) Raises: ValueError (If neither window nor (frac, x) are provided.)