Method Cards

This page is generated from MethodRegistry.list_catalog() so the human-facing method cards stay aligned with the machine-facing catalog contract.

Current package version target: 0.1.1.

Source citations and official upstream package links are collected in Method References.

This page intentionally keeps cards compact. Use Method Matrix for table comparison and Config Reference for full parameter semantics.

The tsdecomp top-level alias remains compatibility-only through 0.1.x and is not the canonical surface for any method listed below.

Flagship methods

MSSA

  • Summary: Multivariate SSA for shared-structure decomposition across channels.
  • Use when: multivariate component recovery; shared seasonal structure across channels
  • Avoid when: single-series workflows where a univariate flagship method is sufficient; very short series that cannot support a sensible window length
  • Key params: window (required), rank (null), primary_period (null)
  • Input/backend: multivariate input, native-backed implementation, maturity flagship
  • Optional dependencies: none
  • Output components: trend, season, residual, components.elementary
  • References: Method References

See Config Reference for the full parameter table.

SSA

  • Summary: Singular spectrum analysis for structured univariate decomposition.
  • Use when: accuracy-first univariate decomposition; component recovery
  • Avoid when: shared-model multivariate decomposition problems; very short series that cannot support a sensible window length
  • Key params: window (required), rank (null), primary_period (null)
  • Input/backend: univariate input, native-backed implementation, maturity flagship
  • Optional dependencies: none
  • Output components: trend, season, residual, components.elementary
  • References: Method References

See Config Reference for the full parameter table.

STD

  • Summary: Fast seasonal-trend decomposition with dispersion-aware diagnostics.
  • Use when: fast seasonal-trend baselines; channelwise multivariate workflows
  • Avoid when: problems that require one shared latent model across channels; series where the dominant period is unknown and cannot be inferred reliably
  • Key params: period (required)
  • Input/backend: channelwise input, native-backed implementation, maturity flagship
  • Optional dependencies: none
  • Output components: trend, season, residual, components.dispersion, components.seasonal_shape
  • References: Method References

See Config Reference for the full parameter table.

STDR

  • Summary: Robust seasonal-trend decomposition for noisier periodic signals.
  • Use when: robust seasonal-trend decomposition; channelwise multivariate workflows
  • Avoid when: problems that require one shared latent model across channels; series where the dominant period is unknown and cannot be inferred reliably
  • Key params: period (required)
  • Input/backend: channelwise input, native-backed implementation, maturity flagship
  • Optional dependencies: none
  • Output components: trend, season, residual, components.dispersion, components.seasonal_shape
  • References: Method References

See Config Reference for the full parameter table.

Stable wrappers and retained methods

CEEMDAN

  • Summary: Noise-assisted EMD variant for more stable IMF extraction.
  • Use when: noise-assisted EMD workflows; adaptive decomposition with improved IMF stability
  • Avoid when: shared-model multivariate decomposition problems
  • Key params: trials (50), noise_width (0.05), primary_period (null)
  • Input/backend: univariate input, python implementation, maturity stable
  • Optional dependencies: PyEMD
  • Output components: trend, season, residual, components.imfs
  • References: Method References

See Config Reference for the full parameter table.

EMD

  • Summary: Empirical mode decomposition under the DeTime result contract.
  • Use when: adaptive decomposition of nonlinear signals; IMF-oriented exploratory analysis
  • Avoid when: shared-model multivariate decomposition problems
  • Key params: n_imfs (null), primary_period (null)
  • Input/backend: univariate input, python implementation, maturity stable
  • Optional dependencies: PyEMD
  • Output components: trend, season, residual, components.imfs
  • References: Method References

See Config Reference for the full parameter table.

MA_BASELINE

  • Summary: Simple moving-average baseline for smoke tests and lightweight workflows.
  • Use when: sanity checks; lightweight baseline decomposition
  • Avoid when: shared-model multivariate decomposition problems
  • Key params: trend_window (7), season_period (null)
  • Input/backend: univariate input, native-backed implementation, maturity stable
  • Optional dependencies: none
  • Output components: trend, season, residual
  • References: Method References

See Config Reference for the full parameter table.

MSTL

  • Summary: Statsmodels MSTL wrapped into the DeTime workflow surface.
  • Use when: multiple seasonalities in univariate data; classical decomposition baselines
  • Avoid when: shared-model multivariate decomposition problems; series where the dominant period is unknown and cannot be inferred reliably
  • Key params: periods (required)
  • Input/backend: univariate input, wrapper implementation, maturity stable
  • Optional dependencies: statsmodels
  • Output components: trend, season, residual, components.seasonal_terms
  • References: Method References

See Config Reference for the full parameter table.

ROBUST_STL

  • Summary: Robust STL-style decomposition wrapped for reproducible workflows.
  • Use when: outlier-prone seasonal-trend baselines; classical robust decomposition
  • Avoid when: shared-model multivariate decomposition problems; series where the dominant period is unknown and cannot be inferred reliably
  • Key params: period (required)
  • Input/backend: univariate input, wrapper implementation, maturity stable
  • Optional dependencies: statsmodels
  • Output components: trend, season, residual
  • References: Method References

See Config Reference for the full parameter table.

STL

  • Summary: Classical STL wrapped into the DeTime workflow contract.
  • Use when: classical seasonal-trend baselines; statsmodels-compatible workflows
  • Avoid when: shared-model multivariate decomposition problems; series where the dominant period is unknown and cannot be inferred reliably
  • Key params: period (required)
  • Input/backend: univariate input, wrapper implementation, maturity stable
  • Optional dependencies: statsmodels
  • Output components: trend, season, residual
  • References: Method References

See Config Reference for the full parameter table.

VMD

  • Summary: Variational mode decomposition integrated into the common workflow layer.
  • Use when: band-limited mode separation; frequency-structured univariate workflows
  • Avoid when: shared-model multivariate decomposition problems
  • Key params: K (4), alpha (2000.0), primary_period (null)
  • Input/backend: univariate input, native-backed implementation, maturity stable
  • Optional dependencies: vmdpy, sktime
  • Output components: trend, season, residual, components.modes
  • References: Method References

See Config Reference for the full parameter table.

WAVELET

  • Summary: Wavelet-based decomposition exposed through the common output contract.
  • Use when: multiscale exploratory analysis; wavelet-style trend and detail separation
  • Avoid when: shared-model multivariate decomposition problems
  • Key params: wavelet ("db4"), level (null)
  • Input/backend: univariate input, wrapper implementation, maturity stable
  • Optional dependencies: PyWavelets
  • Output components: trend, season, residual, components.coefficients
  • References: Method References

See Config Reference for the full parameter table.

Optional backend methods

MEMD

  • Summary: Optional multivariate EMD backend for shared oscillatory structure.
  • Use when: multivariate adaptive decomposition; shared oscillatory modes across channels
  • Avoid when: single-series workflows where a univariate flagship method is sufficient; environments where optional backend dependencies cannot be installed
  • Key params: primary_period (null)
  • Input/backend: multivariate input, optional-backend implementation, maturity optional-backend
  • Optional dependencies: PySDKit
  • Output components: trend, season, residual, components.imfs
  • References: Method References

See Config Reference for the full parameter table.

MVMD

  • Summary: Optional multivariate VMD backend for shared frequency structure.
  • Use when: multivariate variational decomposition; shared frequency structure across channels
  • Avoid when: single-series workflows where a univariate flagship method is sufficient; environments where optional backend dependencies cannot be installed
  • Key params: K (4), alpha (2000.0), primary_period (null)
  • Input/backend: multivariate input, optional-backend implementation, maturity optional-backend
  • Optional dependencies: PySDKit
  • Output components: trend, season, residual, components.modes
  • References: Method References

See Config Reference for the full parameter table.

Experimental methods

AMD_BLOCK

  • Summary: AMD-inspired multiscale smoothing head with periodic-template seasonal reconstruction.
  • Use when: multiscale neural decomposition comparisons; seasonal signals where multiple smoothing scales are informative
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: primary_period (null), fit_scope ("full"), multiscale_windows (null)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

AUTOFORMER_BLOCK

  • Summary: Standalone moving-average decomposition head extracted from the Autoformer architecture.
  • Use when: neural-architecture-inspired seasonal-trend baselines; Autoformer-style decomposition ablations
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: moving_avg (null), primary_period (null)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.moving_mean
  • References: Method References

See Config Reference for the full parameter table.

DELELSTM_BLOCK

  • Summary: DeLELSTM-inspired Holt trend plus periodic-template seasonal decomposition head.
  • Use when: LSTM decomposition-head ablations; signals with smooth level and slope structure
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: primary_period (null), fit_scope ("full"), alpha (0.4), beta (0.2)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

DLINEAR_BLOCK

  • Summary: Standalone moving-average decomposition head extracted from DLinear-style forecasting blocks.
  • Use when: DLinear-style trend/season split baselines; fast neural decomposition head comparisons
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: moving_avg (null), primary_period (null)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.moving_mean
  • References: Method References

See Config Reference for the full parameter table.

FREQMOE_BLOCK

  • Summary: FreqMoE-inspired frequency mixture head for trend plus multi-band seasonal reconstruction.
  • Use when: frequency-mixture neural head ablations; multi-band seasonal decomposition experiments
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: primary_period (null), fit_scope ("full"), trend_window (null), num_bands (4)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

GABOR_CLUSTER

  • Summary: Experimental clustering-based decomposition path.
  • Use when: research prototypes; exploratory clustering-style decomposition
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: model (null), model_path (null)
  • Input/backend: univariate input, native-backed implementation, maturity experimental
  • Optional dependencies: faiss
  • Output components: trend, season, residual, components.clusters
  • References: Method References

See Config Reference for the full parameter table.

INPARFORMER_BLOCK

  • Summary: InParformer-inspired moving-average trend plus periodic-template seasonal decomposition head.
  • Use when: periodic-template neural decomposition baselines; prefix/full-scope ablation experiments
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: primary_period (null), fit_scope ("full"), trend_window (null)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

LEDDAM_BLOCK

  • Summary: LEDDAM LD smoothing block exposed as a Gaussian-kernel decomposition operator.
  • Use when: LEDDAM-style decomposition ablations; kernel smoothing neural head comparisons
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: kernel_size (25), sigma (1.0)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.ld_trend, components.kernel
  • References: Method References

See Config Reference for the full parameter table.

MOVING_AVERAGE_DECOMPOSITION_BLOCK

  • Summary: Generic neural forecasting moving-average decomposition block exposed as a DeTime method.
  • Use when: generic decomposition-block smoke tests; fast moving-average neural head baselines
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: moving_avg (null), primary_period (null)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.moving_mean
  • References: Method References

See Config Reference for the full parameter table.

NBEATS_INTERPRETABLE

  • Summary: Torch-backed interpretable N-BEATS trend and seasonality stacks used as a learned decomposition prior.
  • Use when: learned-basis decomposition experiments; N-BEATS interpretable-stack ablations
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: degree_of_polynomial (3), num_harmonics (8), fit_scope ("full"), n_epochs (200)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: torch
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

PARSIMONY_BLOCK

  • Summary: Parsimony-inspired trend head with compact harmonic seasonal projection.
  • Use when: compact harmonic decomposition baselines; low-parameter neural head comparisons
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: primary_period (null), fit_scope ("full"), trend_window (null), num_harmonics (2)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

ST_MTM_BLOCK

  • Summary: ST-MTM-inspired smoothing head combining trend smoothing and smoothed periodic seasonality.
  • Use when: seasonal-trend pretraining block ablations; smooth periodic decomposition baselines
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: primary_period (null), fit_scope ("full"), trend_window (null), season_smooth_window (null)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

TIMEKAN_BLOCK

  • Summary: TimeKAN-inspired decomposition head blending template and harmonic seasonal estimates.
  • Use when: KAN-inspired neural decomposition ablations; frequency-template hybrid seasonal baselines
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: primary_period (null), fit_scope ("full"), trend_window (null), num_bands (2)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

TIMES2D_BLOCK

  • Summary: Times2D-inspired multi-period harmonic decomposition head.
  • Use when: multi-period neural decomposition baselines; FFT-selected seasonal period comparisons
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: primary_period (null), fit_scope ("full"), top_k_periods (2), num_harmonics (1), trend_window (null)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

WAVEFORM_BLOCK

  • Summary: WaveForM-inspired wavelet multiresolution decomposition head.
  • Use when: wavelet neural-head ablations; multiresolution trend/detail comparisons
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: wavelet ("db4"), level (3), season_levels ([1, 2])
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: PyWavelets
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

WAVELETMIXER_BLOCK

  • Summary: WaveletMixer-inspired multiresolution decomposition head using mixed wavelet detail levels.
  • Use when: wavelet-mixer neural decomposition baselines; multi-level detail seasonal reconstruction
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: wavelet ("sym4"), level (4), season_levels ([1, 2, 3])
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: PyWavelets
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.

XPATCH_BLOCK

  • Summary: xPatch-inspired exponential smoothing head for standalone trend and local-season decomposition.
  • Use when: exponential smoothing neural head comparisons; fast local seasonal-trend decomposition
  • Avoid when: shared-model multivariate decomposition problems; first-pass baselines or high-trust production workflows
  • Key params: ma_type ("ema"), trend_window (null), season_smooth (null)
  • Input/backend: univariate input, python implementation, maturity experimental
  • Optional dependencies: none
  • Output components: trend, season, residual, components.trend, components.season
  • References: Method References

See Config Reference for the full parameter table.