tsbootstrap

User guide

  • What’s new in 0.3.0
    • Compiled backend (backend="compiled")
      • Installation
      • When to use it
      • Reproducibility note
      • Example: bootstrap() with the compiled backend
      • Example: bootstrap_reduce() with the compiled backend
    • Named reducers for bootstrap_reduce() and bootstrap_reduce_panel()
      • Available named reducers
      • Callable statistic (numpy backend only)
      • Example: named reducers
    • float32 simulation dtype (dtype="float32")
    • bootstrap_reduce_panel(): whole-panel calibration
      • Signature
      • Input forms
      • Output shape
      • Reproducibility note
      • Supported methods
      • Backend
    • Quick-reference table
  • Quickstart
    • Installation
    • Basic usage
    • Choosing a method
    • Inputs
    • Reproducibility
    • sktime ecosystem
  • Method specifications
    • Observation-resampling methods
      • IID
      • MovingBlock
      • CircularBlock
      • StationaryBlock
      • NonOverlappingBlock
      • TaperedBlock
    • Model-based methods
      • ResidualBootstrap
      • SieveAR
    • Deferred methods
  • Results
    • BootstrapResult
      • Accessing samples
      • Out-of-bag / in-bag primitives
    • BootstrapSample
    • BootstrapRunMetadata
      • Handling failed runs
  • Diagnostics
    • The Diagnosis object
    • Interpretation
    • Example workflow
  • Uncertainty quantification guide
    • Regression prediction intervals: EnbPI
    • Calibrators: choosing the half-width
    • Forecast intervals
    • Scaling calibration: bootstrap_reduce
  • sktime / skbase adapters
    • Installation
    • Using an adapter
    • Available adapters
      • Block method adapters
      • Model method adapters
    • Validating with sktime
    • Relationship to the functional API

Tutorials

  • Tutorials
    • Know your data: datasets and synthetic processes
      • Imports
      • Bundled real datasets
      • Synthetic data-generating processes
      • A diagnose() preview
    • Quickstart: your first bootstrap
      • A small series
      • One bootstrap call
      • Going faster (optional)
      • Out-of-bag mask
      • Which method should I use?
      • Visualizing replicates
    • Which bootstrap should I use?
      • Helpers: DGPs, an OLS lag-1 estimator, and a coverage Monte-Carlo
      • Recipe 1: IID resampling destroys autocorrelation
      • Recipe 2: block edge effects and the downward AR-coefficient bias
      • Recipe 3: block length is a bias-variance knob
      • Recipe 4: model-based bootstraps under misspecification
      • Capstone: diagnose() as an advisor
    • Cheat sheet: pick your bootstrap
      • The capability table
      • The decision tree
      • Where to go next
    • Observation resampling: the block-method family
      • Two test series
      • A helper to plot replicates
      • 1. IID
      • 2. MovingBlock
      • 3. CircularBlock
      • 4. StationaryBlock
      • 5. NonOverlappingBlock
      • 6. TaperedBlock
      • Edge effects: MovingBlock starves the ends, CircularBlock is flat
      • Automatic block length: the Politis-White rule
    • Model-based bootstraps: residual and sieve
      • A synthetic AR(2)
      • ResidualBootstrap with a fixed-order AR
      • ResidualBootstrap with ARIMA on a near-unit-root series
      • SieveAR: let the data choose the order
      • The stability guard
      • The three model-based specs at a glance
    • Multivariate bootstrap: keeping series in step
      • A synthetic bivariate VAR(1) with known structure
      • Two kinds of cross-correlation
      • VAR residual bootstrap
      • Block methods on multivariate input
      • Where the naive choice fails
      • A real macro example
      • diagnose() on multivariate input
    • EnbPI prediction intervals and the four calibrators
      • Inline data-generating processes and helpers
      • A regression-from-lags problem
      • All four calibrators from one fit
      • Interval-quality plot 1: the interval over time, with misses highlighted
      • Interval-quality plot 2: running coverage versus the nominal line
      • Interval-quality plot 3: interval width over time
      • Interval-quality plot 4: coverage versus target as alpha sweeps
      • The stress test: a GARCH(1,1) volatility-clustering series
      • GARCH plot: width tracking the true conditional volatility
      • GARCH plot: running coverage separates the calibrators
      • Interval-quality plot 5: conditional coverage by volatility bin
      • Winkler scores: one number to compare calibrators
      • What to take away
    • Adaptive conformal under distribution shift
      • What each method actually guarantees
      • The static interval mis-covers across the break
      • ACI re-calibrates online and recovers nominal coverage
      • Rolling coverage: static dips, ACI tracks
      • What ACI did to the level
      • Static versus ACI, split at the break
      • The rolling picture and the half-widths
      • A single snapshot: decay = 1 versus decay < 1
      • Online coverage: ordinary quantile under-covers the new regime
      • The recency-weighting trade, plotted
    • Forward forecast intervals
      • How a forward interval is built
      • Data and helpers
      • An AR forecast fan on the lynx series
      • Does the band cover at the nominal rate?
      • Uncertainty grows with the horizon
      • Supported models: AR only
      • The forecast fan in one paragraph
    • sktime / skbase adapters: bootstraps as estimator classes
      • The adapter classes
      • The data: airline passengers
      • Constructing an estimator and running its generator
      • Three adapters end to end
      • return_indices: where each value came from
      • sktime tags
      • The skbase object contract
      • When to reach for the adapters
    • Bootstrapping with exogenous regressors (ARX / VARX / ARIMAX)
      • What accepts exog, and what does not
      • ARX: recover a known covariate effect
      • VARX: a multivariate covariate effect
      • ARIMAX: a covariate on a trending series
      • When exog is rejected: a non-model method
      • Working with covariates: the rules
    • Reproducibility and memory-bounded scaling
      • A small AR(1) series
      • Reproducibility: same seed, identical results
      • Provenance: the run records how it was produced
      • Memory-bounded scaling to large B
      • A confidence interval without the full array
      • The reduced CI matches the full-array CI
      • Putting it together
      • Closing the loop on reproducibility
      • Scaling across many series at once

API reference

  • bootstrap()
    • bootstrap()
  • Method specifications (tsbootstrap.methods)
    • BaseMethodSpec
      • BaseMethodSpec.model_config
    • IID
      • IID.model_config
    • MovingBlock
      • MovingBlock.model_config
    • CircularBlock
      • CircularBlock.model_config
    • StationaryBlock
      • StationaryBlock.model_config
    • NonOverlappingBlock
      • NonOverlappingBlock.model_config
    • TaperedBlock
      • TaperedBlock.model_config
    • AR
      • AR.model_config
    • ARIMA
      • ARIMA.model_config
    • VAR
      • VAR.model_config
    • ResidualBootstrap
      • ResidualBootstrap.model_config
    • SieveAR
      • SieveAR.model_config
    • OBSERVATION_RESAMPLING
  • Results (tsbootstrap.results)
    • BootstrapRunMetadata
    • BootstrapSample
      • BootstrapSample.values
      • BootstrapSample.sample_id
      • BootstrapSample.indices
      • BootstrapSample.metadata
    • BootstrapResult
      • BootstrapResult.iter_samples()
      • BootstrapResult.values()
      • BootstrapResult.indices()
      • BootstrapResult.inbag_counts()
      • BootstrapResult.get_oob_mask()
    • ReducedResult
      • ReducedResult.failed
      • ReducedResult.failure_reason
      • ReducedResult.quantile()
  • Diagnostics (tsbootstrap.diagnostics)
    • Diagnosis
    • diagnose()
  • Uncertainty quantification (tsbootstrap.uq)
    • EnbPIEnsemble
      • EnbPIEnsemble.fit()
      • EnbPIEnsemble.oob_residuals
      • EnbPIEnsemble.oob_prediction
      • EnbPIEnsemble.predict_interval()
    • fit_predict_oob()
    • enbpi_intervals()
    • forecast_intervals()
    • aci_halfwidths()
    • nexcp_quantile()
    • static_halfwidths()
    • sliding_window_halfwidths()
  • sktime adapters (tsbootstrap.adapters)
    • BaseTimeSeriesBootstrap
      • BaseTimeSeriesBootstrap.bootstrap()
      • BaseTimeSeriesBootstrap.get_n_bootstraps()
      • BaseTimeSeriesBootstrap.get_test_params()
    • IIDBootstrap
      • IIDBootstrap.get_test_params()
    • MovingBlockBootstrap
      • MovingBlockBootstrap.get_test_params()
    • CircularBlockBootstrap
      • CircularBlockBootstrap.get_test_params()
    • StationaryBlockBootstrap
      • StationaryBlockBootstrap.get_test_params()
    • NonOverlappingBlockBootstrap
      • NonOverlappingBlockBootstrap.get_test_params()
    • TaperedBlockBootstrap
      • TaperedBlockBootstrap.get_test_params()
    • ARResidualBootstrap
      • ARResidualBootstrap.get_test_params()
    • ARIMAResidualBootstrap
      • ARIMAResidualBootstrap.get_test_params()
    • VARResidualBootstrap
      • VARResidualBootstrap.get_test_params()
    • SieveBootstrap
      • SieveBootstrap.get_test_params()
    • BaseTimeSeriesBootstrap
      • BaseTimeSeriesBootstrap.bootstrap()
      • BaseTimeSeriesBootstrap.get_n_bootstraps()
      • BaseTimeSeriesBootstrap.get_test_params()
    • IIDBootstrap
      • IIDBootstrap.get_test_params()
    • MovingBlockBootstrap
      • MovingBlockBootstrap.get_test_params()
    • CircularBlockBootstrap
      • CircularBlockBootstrap.get_test_params()
    • StationaryBlockBootstrap
      • StationaryBlockBootstrap.get_test_params()
    • NonOverlappingBlockBootstrap
      • NonOverlappingBlockBootstrap.get_test_params()
    • TaperedBlockBootstrap
      • TaperedBlockBootstrap.get_test_params()
    • ARResidualBootstrap
      • ARResidualBootstrap.get_test_params()
    • ARIMAResidualBootstrap
      • ARIMAResidualBootstrap.get_test_params()
    • VARResidualBootstrap
      • VARResidualBootstrap.get_test_params()
    • SieveBootstrap
      • SieveBootstrap.get_test_params()
  • Errors and warnings (tsbootstrap.errors)
    • Codes
    • TSBootstrapError
    • InputDataError
    • MethodConfigError
    • ModelStabilityError
    • BackendError
    • OOBUnavailableError
    • RNGContractError
    • TSBootstrapWarning
    • NearUnitRootWarning
    • DegenerateBlockBootstrapWarning
tsbootstrap
  • Python Module Index

Python Module Index

t
 
t
- tsbootstrap
    tsbootstrap.adapters
    tsbootstrap.adapters.estimators
    tsbootstrap.diagnostics
    tsbootstrap.errors
    tsbootstrap.methods
    tsbootstrap.results
    tsbootstrap.uq

© Copyright 2023 - 2026 (MIT License), Sankalp Gilda.

Built with Sphinx using a theme provided by Read the Docs.