API reference¶

Generated from the source, so it cannot drift from what the package actually exports.

Estimating a trend¶

The functional surface. Each builds a smoother, fits it, and returns a DataFrame carrying the columns described in Uncertainty.

`naive_trend`	Estimate the trend by central finite differences.
`sgolay_trend`	Estimate the trend with a Savitzky-Golay filter.
`spline_trend`	Estimate the trend with a knot-selecting smoothing spline.
`pspline_trend`	Estimate the trend with a penalized smoothing spline.
`loess_trend`	Estimate the trend with LOESS.
`local_polynomial_trend`	Estimate the trend by local polynomial regression.
`l1_trend_filter`	Estimate a piecewise-polynomial trend with sparse changes in slope.
`gp_trend`	Estimate the trend with Gaussian process regression.
`kalman_trend`	Estimate the trend with a local linear trend state-space model.
`estimate_trend`	Estimate a trend, choosing the method automatically if asked.
`select_trend_method`	Recommend a smoother for a series.
`estimate`	Run a smoother over a DataFrame and return the structured estimate.

Smoothers¶

The objects behind those functions. Use these directly when you want to hand one to SiZer, to trend_with_deseasonalization, or to estimate.

Whether a smoother is linear decides how its uncertainty is computed: linear smoothers get an exact operator variance, the rest are bootstrapped.

`Smoother`	Base class for every trend estimator.
`SavitzkyGolay`	Savitzky-Golay filter: local polynomial least squares on a fixed window.
`NaiveDifference`	Central finite differences on the raw series.
`LocalPolynomial`	Local polynomial regression with kernel weights.
`Loess`	LOESS smoothing, then a local polynomial fit to the smooth.
`PenalizedSpline`	Smoothing spline with a fixed roughness penalty.
`InterpolatingSpline`	Smoothing spline that chooses its own knots.
`L1TrendFilter`	L1 trend filtering: piecewise-polynomial fit with sparse kinks.
`GaussianProcess`	Gaussian process regression with an exact derivative posterior.
`StateSpace`	Local linear trend, optionally with seasonal and damping.
`build`	Construct a smoother by name.

incline.SMOOTHERS is the registry those names are looked up in: a mapping from name to class. build(name, **kwargs) constructs from it, and estimate_trend(method=...) dispatches through it, so a newly registered smoother is reachable everywhere without editing a dispatch table.

Noise models¶

What noise= accepts. The default assumes independence; under real autocorrelation that understates the uncertainty substantially.

`NoiseModel`	Base class for noise specifications.
`IID`	Independent noise with constant variance.
`AR1`	First-order autoregressive noise.
`Heteroskedastic`	Independent noise whose scale changes across the series.
`Given`	A caller-supplied noise covariance, used exactly as provided.
`local_sigma`	Estimate a noise level that varies across the series.

Results¶

`TrendEstimate`	A smoothed series and the derivative of that smooth, with uncertainty.
`TimeAxis`	Numeric time positions for a series.

Multi-scale analysis¶

Which features survive being looked at from every smoothing scale.

`SiZer`	A scale sweep over any smoother.
`SiZerMap`	The result of a scale sweep.
`sizer_analysis`	Run a scale sweep in one call.
`trend_with_sizer`	Estimate a trend and mark where it survives a scale sweep.

Seasonality¶

Decomposition is preprocessing: deseasonalize returns a frame, which any estimator then accepts.

`deseasonalize`	Split a series into trend, cycle and remainder.
`detect_seasonality`	Look for a repeating cycle, by three methods in decreasing reliability.
`stl_decompose`	Seasonal-trend decomposition by LOESS.
`moving_average_decompose`	Classical decomposition by a centered moving average.
`trend_with_deseasonalization`	Deseasonalize, then estimate the trend of what is left.
`Seasonality`	What was found when looking for a cycle.

Ranking many series¶

trending

Rank series by recent trend strength.

Simulation¶

Synthetic series with known derivatives, used by the package’s own calibration tests and available for yours.

`generate_time_series`	Simulate a series from a known trend.
`standard_test_functions`	A spread of trends covering the shapes estimators handle differently.