Method Comparison Matrix

This page is generated from MethodRegistry.list_catalog() and summarizes method-level behavior for onboarding, review, and machine-facing routing.

Current package version target: 0.1.1.

Method Input mode Backend Maturity Required/common params Optional deps Native Multivariate Output components Recommended use
AMD_BLOCK univariate python experimental primary_period (null), fit_scope ("full"), multiscale_windows (null) none no univariate trend, season, residual, components.trend, components.season multiscale neural decomposition comparisons; seasonal signals where multiple smoothing scales are informative
AUTOFORMER_BLOCK univariate python experimental moving_avg (null), primary_period (null) none no univariate trend, season, residual, components.moving_mean neural-architecture-inspired seasonal-trend baselines; Autoformer-style decomposition ablations
CEEMDAN univariate python stable trials (50), noise_width (0.05), primary_period (null) PyEMD no univariate trend, season, residual, components.imfs noise-assisted EMD workflows; adaptive decomposition with improved IMF stability
DELELSTM_BLOCK univariate python experimental primary_period (null), fit_scope ("full"), alpha (0.4), beta (0.2) none no univariate trend, season, residual, components.trend, components.season LSTM decomposition-head ablations; signals with smooth level and slope structure
DLINEAR_BLOCK univariate python experimental moving_avg (null), primary_period (null) none no univariate trend, season, residual, components.moving_mean DLinear-style trend/season split baselines; fast neural decomposition head comparisons
EMD univariate python stable n_imfs (null), primary_period (null) PyEMD no univariate trend, season, residual, components.imfs adaptive decomposition of nonlinear signals; IMF-oriented exploratory analysis
FREQMOE_BLOCK univariate python experimental primary_period (null), fit_scope ("full"), trend_window (null), num_bands (4) none no univariate trend, season, residual, components.trend, components.season frequency-mixture neural head ablations; multi-band seasonal decomposition experiments
GABOR_CLUSTER univariate native-backed experimental model (null), model_path (null) faiss yes univariate trend, season, residual, components.clusters research prototypes; exploratory clustering-style decomposition
INPARFORMER_BLOCK univariate python experimental primary_period (null), fit_scope ("full"), trend_window (null) none no univariate trend, season, residual, components.trend, components.season periodic-template neural decomposition baselines; prefix/full-scope ablation experiments
LEDDAM_BLOCK univariate python experimental kernel_size (25), sigma (1.0) none no univariate trend, season, residual, components.ld_trend, components.kernel LEDDAM-style decomposition ablations; kernel smoothing neural head comparisons
MA_BASELINE univariate native-backed stable trend_window (7), season_period (null) none yes univariate trend, season, residual sanity checks; lightweight baseline decomposition
MEMD multivariate optional-backend optional-backend primary_period (null) PySDKit no shared-model trend, season, residual, components.imfs multivariate adaptive decomposition; shared oscillatory modes across channels
MOVING_AVERAGE_DECOMPOSITION_BLOCK univariate python experimental moving_avg (null), primary_period (null) none no univariate trend, season, residual, components.moving_mean generic decomposition-block smoke tests; fast moving-average neural head baselines
MSSA multivariate native-backed flagship window (required), rank (null), primary_period (null) none yes shared-model trend, season, residual, components.elementary multivariate component recovery; shared seasonal structure across channels
MSTL univariate wrapper stable periods (required) statsmodels no univariate trend, season, residual, components.seasonal_terms multiple seasonalities in univariate data; classical decomposition baselines
MVMD multivariate optional-backend optional-backend K (4), alpha (2000.0), primary_period (null) PySDKit no shared-model trend, season, residual, components.modes multivariate variational decomposition; shared frequency structure across channels
NBEATS_INTERPRETABLE univariate python experimental degree_of_polynomial (3), num_harmonics (8), fit_scope ("full"), n_epochs (200) torch no univariate trend, season, residual, components.trend, components.season learned-basis decomposition experiments; N-BEATS interpretable-stack ablations
PARSIMONY_BLOCK univariate python experimental primary_period (null), fit_scope ("full"), trend_window (null), num_harmonics (2) none no univariate trend, season, residual, components.trend, components.season compact harmonic decomposition baselines; low-parameter neural head comparisons
ROBUST_STL univariate wrapper stable period (required) statsmodels no univariate trend, season, residual outlier-prone seasonal-trend baselines; classical robust decomposition
SSA univariate native-backed flagship window (required), rank (null), primary_period (null) none yes univariate trend, season, residual, components.elementary accuracy-first univariate decomposition; component recovery
STD channelwise native-backed flagship period (required) none yes channelwise trend, season, residual, components.dispersion, components.seasonal_shape fast seasonal-trend baselines; channelwise multivariate workflows
STDR channelwise native-backed flagship period (required) none yes channelwise trend, season, residual, components.dispersion, components.seasonal_shape robust seasonal-trend decomposition; channelwise multivariate workflows
STL univariate wrapper stable period (required) statsmodels no univariate trend, season, residual classical seasonal-trend baselines; statsmodels-compatible workflows
ST_MTM_BLOCK univariate python experimental primary_period (null), fit_scope ("full"), trend_window (null), season_smooth_window (null) none no univariate trend, season, residual, components.trend, components.season seasonal-trend pretraining block ablations; smooth periodic decomposition baselines
TIMEKAN_BLOCK univariate python experimental primary_period (null), fit_scope ("full"), trend_window (null), num_bands (2) none no univariate trend, season, residual, components.trend, components.season KAN-inspired neural decomposition ablations; frequency-template hybrid seasonal baselines
TIMES2D_BLOCK univariate python experimental primary_period (null), fit_scope ("full"), top_k_periods (2), num_harmonics (1), trend_window (null) none no univariate trend, season, residual, components.trend, components.season multi-period neural decomposition baselines; FFT-selected seasonal period comparisons
VMD univariate native-backed stable K (4), alpha (2000.0), primary_period (null) vmdpy, sktime yes univariate trend, season, residual, components.modes band-limited mode separation; frequency-structured univariate workflows
WAVEFORM_BLOCK univariate python experimental wavelet ("db4"), level (3), season_levels ([1, 2]) PyWavelets no univariate trend, season, residual, components.trend, components.season wavelet neural-head ablations; multiresolution trend/detail comparisons
WAVELET univariate wrapper stable wavelet ("db4"), level (null) PyWavelets no univariate trend, season, residual, components.coefficients multiscale exploratory analysis; wavelet-style trend and detail separation
WAVELETMIXER_BLOCK univariate python experimental wavelet ("sym4"), level (4), season_levels ([1, 2, 3]) PyWavelets no univariate trend, season, residual, components.trend, components.season wavelet-mixer neural decomposition baselines; multi-level detail seasonal reconstruction
XPATCH_BLOCK univariate python experimental ma_type ("ema"), trend_window (null), season_smooth (null) none no univariate trend, season, residual, components.trend, components.season exponential smoothing neural head comparisons; fast local seasonal-trend decomposition

Use Config Reference for full DecompositionConfig field semantics and per-method parameter descriptions.

Use Method References for primary literature and official upstream package links.