tsbootstrap: v0.3.0

0.3.0 (2026-06-23)

Features

• scikit-base registry and testing framework (#26) (4fc838e)
• skbase inheritance for config objects (#31) (24ced3c)
• skbase inheritance in central base classes (#10) (3d650b8)
• 44 sklearn compatible interface (with fkiraly) (#66) (1b521ca)
• accept pandas/Polars DataFrames via the narwhals boundary (c4fd180)
• adaptive and nonexchangeable conformal calibration (28396c7)
• add compiled acceleration backend, named reducers, float32 dtype, and panel reduce (077a258)
• added dunder methods (8d98867)
• ARIMAX exogenous-covariate support (regression with ARIMA errors) (5fb96e1)
• block index kernels and automatic block length (59c169f)
• bootstrap_reduce for streaming per-replicate statistics (64f30d8)
• change bootstrap test_size argument default to 0.0 (#21) (152f0fd)
• created separate config module (723b3c8)
• diagnose() method advisor (8928be9)
• enable backends by default with 7.66x performance improvement (#196) (8b5e410), closes #194
• ensure bootstrap tests also run via check_estimator (#84) (a97ee34)
• exogenous-covariate support for the AR residual bootstrap (e44328c)
• extend suite test coverage to further bootstrap contract elements - equal shape as origial series, test_ratio, number of bootstraps (#83) (0c15713)
• isolate arch and statsmodels as soft dependencies, new all_extras soft dependency set (#39) (a372c1a)
• isolate scikit-learn-extra (#52) (5e4cce3)
• mcp: add a read-only MCP server with two bootstrap tools (07bb9cd)
• re-export the uncertainty-quantification surface at the top level (7b2f623)
• recursive AR and sieve bootstrap engine (55d9e81)
• recursive ARIMA bootstrap on the differenced scale (7b8fcec)
• recursive VAR bootstrap (multivariate) (38277a7)
• remove numba (#24) (5a8d595)
• removes stray todo comments from base_bootstrap (#101) (4da06ec)
• separated ranklags into a module (27ee6ea)
• skbase/sktime bootstrap adapters (concrete over shared base) (818925e)
• stability_policy=skip for pipeline-survivable failed fits (fa2d5e8)
• structured errors, deterministic RNG contract, input validation (4741a6b)
• tapered block bootstrap with energy-normalized windows (dd23d6d)
• typed method API, structured results, i.i.d. baseline (d5a31fc)
• UQ layer with EnbPI and bootstrap forecast intervals (175724d)
• uq: EnbPIEnsemble fit/predict object with calibrator family and sliding-window EnbPI (965074a)
• VARX exogenous-covariate support for the VAR residual bootstrap (5ba9c43)

Bug Fixes

• added tsbs logo (289087f)
• ARIMA residuals consistent with the lfilter simulation engine (070faab)
• avoid float-equality checks flagged by SonarCloud (c0b2a6e)
• change exog to y in methods of BaseTimeSeriesBootstrap (#30) (#32) (b3f6053)
• change exog to y in methods of BaseTimeSeriesBootstrap (#30) (#33) (a5aba98)
• condition the ARIMA bootstrap on the observed initial state (4fd51c1)
• ensure check_estimator runs bootstrap tests for bootstraps (#87) (ba18672)
• failing tests (202f93f)
• PR CI trigger condition (#11) (d2862fb)
• quality: clear SonarCloud findings and scope analysis to src (517cfc2)
• remove fail-fast condition for matrix CI (#19) (2a12437)
• removed erraneously comments out test code (6cea59b)
• run each warm-up hook once regardless of registration order (3349f13)
• temp commented out residual testing (f8889cb)
• temporarily not publishing coverage reports (2037b71)
• temporarily skip sporadically failing tests due to LU decomposition (#28) (7f7d357)

Performance Improvements

• add a profiling harness for the hot paths (71da5bc)
• add asv benchmark suite for the bootstrap engines (347e3b5)
• compiled replicate-parallel VAR kernel behind the accel extra (67f5b8a)
• fit AR/VAR/sieve by direct OLS instead of statsmodels (f5962fa)
• true-vectorize the recursive engines across replicates (9c46c03)
• vectorize the AR initial-state across paths (7d803e5)

Documentation

• add a minimal AGENTS.md for coding agents (3f18347)
• add an executable notebook gallery with the quickstart tutorial (4624085)
• add Context7 and DeepWiki indexing (95f9df5)
• add core-methods tutorial notebooks (312834c)
• add decision log and a detailed ARIMAX/VARX plan (9186627)
• add getting-started and decision-guide tutorial notebooks (928c840)
• add the uncertainty-quantification guide and API reference (3d6492f)
• add uncertainty-quantification and integration tutorial notebooks (c7ed2e6)
• add v0.2.0 plan and development backlog (45a58d9)
• added readthedocsyaml (8bb0001)
• bump CITATION.cff version to 0.2.0 (3bbdbef)
• capture future-capability roadmap and the exog scope decision (3285ad0)
• correct stale guides and changelog against the v0.2.0 code (afa60af)
• docstring for bootstrap method in base class (#20) (2a50e20)
• fixed error with tukeyalpha (46f3ccd)
• for new src files (3ee4f1b)
• humanize README and CONTRIBUTING prose (e64367b)
• installed tsbootstrap in readthedocs, enabled build pull requests (#72) (16db5c1)
• mark exogenous support (ARX/VARX/ARIMAX) complete (0b38699)
• mcp: add the registry manifest, README section, and publish workflow (44ede19)
• remove the Code Climate maintainability badge (b793157)
• replace em dashes with plain punctuation across docs, docstrings, and comments (8641a32)
• rewrite README for the typed v0.2.0 API (14af6a0)
• rewrite the Sphinx documentation for the new API (511ed4d)
• state behaviour directly in docstrings and notes (3b866e6)
• tighten the contributor and migration guides and align the roadmap (c3db0f6)
• updated contributing guide (#78) (5edadc7)
• updated docstrings (9e9df1f)
• updated pyproject.toml dependencies and .readthedocs.yaml (653abb9)
• wire the tutorial gallery and refresh the README for v0.2.0 (7f7ea62)