smooth documentation

Classes

ADAM([model, lags, ar_order, i_order, ...])	ADAM: Augmented Dynamic Adaptive Model for Time Series Forecasting.
ES([model, lags, persistence, phi, initial, ...])	Exponential Smoothing in Single Source of Error (SSOE) state space model.

ADAM Methods

ADAM.fit(y[, X])	Fit the ADAM model to time series data.
ADAM.predict(h[, X, calculate_intervals, ...])	Generate point forecasts using the fitted ADAM model.
ADAM.predict_intervals(h[, X, levels, side])	Generate prediction intervals using the fitted ADAM model.
ADAM.select_best_model()	Select the best model based on information criteria and update model parameters.
ADAM.summary([digits])	Generate a formatted summary of the fitted model.

ES Methods

ES.fit(y[, X])	Fit the ADAM model to time series data.
ES.predict(h[, X, calculate_intervals, ...])	Generate point forecasts using the fitted ADAM model.
ES.predict_intervals(h[, X, levels, side])	Generate prediction intervals using the fitted ADAM model.
ES.select_best_model()	Select the best model based on information criteria and update model parameters.
ES.summary([digits])	Generate a formatted summary of the fitted model.

Utility Functions

• msdecompose - Multiple seasonal decomposition for time series

• lowess - LOWESS (Locally Weighted Scatterplot Smoothing)

Optimization Settings

The ADAM and ES classes use the NLopt library for parameter optimization. You can customize the optimization behavior via the nlopt_kargs parameter:

from smooth import ADAM

model = ADAM(
    model="AAN",
    nlopt_kargs={
        "print_level": 1,        # Print optimization progress
        "xtol_rel": 1e-8,        # Relative parameter tolerance
        "algorithm": "NLOPT_LN_SBPLX"  # Use Subplex algorithm
    }
)
model.fit(y)

Available parameters:

Parameter	Description	Default
print_level	Verbosity level. When >0, prints B and CF on every iteration.	0
xtol_rel	Relative tolerance on parameters. Stops when changes < xtol_rel * |params|.	1e-6
xtol_abs	Absolute tolerance on parameters. Stops when changes < xtol_abs.	1e-8
ftol_rel	Relative tolerance on cost function. Stops when changes < ftol_rel * |CF|.	1e-8
ftol_abs	Absolute tolerance on cost function. Stops when changes < ftol_abs.	0
algorithm	NLopt algorithm name. Use “LN_” prefix for derivative-free. Options: NLOPT_LN_NELDERMEAD, NLOPT_LN_SBPLX, NLOPT_LN_COBYLA, NLOPT_LN_BOBYQA.	NLOPT_LN_NELDERMEAD