Bump pyvinecopulib from 0.7.6 to 1.0.0 in /examples

[dependabot]

Bumps pyvinecopulib from 0.7.6 to 1.0.0.

Release notes

Sourced from pyvinecopulib's releases.

pyvinecopulib 1.0.0

1.0.0 (2026-09-18)

This release ships the whole vine-modeling stack rather than the copula half of it. Vinedist and the new pyvinecopulib.margins layer put Sklar's theorem on the data scale; pyvinecopulib.torch runs the evaluation cascade on the GPU under autograd; pyvinecopulib.sklearn adds scikit-learn-compatible estimators; and the array-agnostic layer in pyvinecopulib.core hosts custom pair copulas and non-simplified vines. The public API is reorganized into the core, families, utils and margins subpackages, with the pre-reorganization top-level names kept as warning aliases until 2.0.

It also advances all three vendored C++ libraries, so nearly every tll and Kde1d fit moves. Read the migration guide in docs/migrating-to-1.0.md before upgrading from 0.7.x.

Breaking API changes in pyvinecopulib

• Use one argument order on every estimator: the observations first, then controls, then keyword-only whatever the object cannot infer. Vinecop.from_data took controls fifth, behind structure, so the call carried over from fit bound a controls object as a structure (#326).

  • pv.Vinecop.from_data(u, structure, matrix, var_types, controls) -> pv.Vinecop.from_data(u, controls, structure=..., matrix=..., var_types=...)
  • pv.Bicop.from_data(u, controls, var_types) -> pv.Bicop.from_data(u, controls, var_types=...)
  • pv.core.Kde1d.fit(x, weights) -> pv.core.Kde1d.fit(y, FitControlsKde1d(weights=...)), and likewise select; the kernel knobs and the weights are all FitControlsKde1d, and the observations are y, because x means exogenous covariates everywhere in the Python API (#339)
  • likewise on BicopBase, VinecopBase, VinedistBase and their torch subclasses; MarginBase already read this way

• Reorganize the public API into the core, families and utils subpackages; the family constants, Kde1d and the utility functions still resolve at the top level, but warn on access and are removed in 2.0 (#207, #292).

  • pyvinecopulib.gaussian -> pyvinecopulib.families.gaussian, and likewise every family constant and group
  • pyvinecopulib.Kde1d -> pyvinecopulib.core.Kde1d
  • pyvinecopulib.wdm -> pyvinecopulib.utils.wdm, and likewise sobol, ghalton, benchmark and pairs_copula_data

• Default selection_criterion to "aic" instead of "bic" on FitControlsBicop and FitControlsVinecop, matching the C++ and R defaults; pass selection_criterion="bic" to keep the previous selection (#251).

• Move every tll fit, and so every unqualified Bicop.from_data / Vinecop.from_data that selects one: the interpolation grid's margins are balanced across both sweep orders and iterated to convergence, and hfunc / hinv no longer floor the interpolated density at 1e-4 (#305, vinecopulib#751).

• Move every Kde1d fit: the plug-in bandwidth shifts by O(1/n), and a bounded fit shifts again under the new endpoint treatment (#312, kde1d#29).

• Fit a finite Kde1d bound with a local-linear boundary estimator by default; pass boundary_repair=False for the previous transformed fit (#312, kde1d#36).

• Read a discrete Kde1d's xmin / xmax as the integer support, so both the bounds and the data must be integers and the fitted grid runs half a unit wider at each end (#312, kde1d#37).

• Label a selected RVineStructure with the conditioned variable on the diagonal, so get_matrix() and order differ for the same model; densities and log-likelihoods do not (#251, vinecopulib#702).

• Rename the sampling entry points to sample, keeping the released spellings as warning aliases (#297).

  • Bicop.simulate(n) -> Bicop.sample(n), and likewise on Vinecop, RVineStructure and Kde1d
  • pyvinecopulib.simulate_uniform -> pyvinecopulib.utils.sample_uniform

• Return self from Bicop.fit / .select and Vinecop.fit / .select instead of None, so every estimator in the package composes the same way (#326).

• Make parameters, num_threads, seeds and randomize_discrete keyword-only on the evaluation methods of Bicop and Vinecop: each class was internally consistent and the two disagreed, Bicop reading (u, parameters, num_threads) where Vinecop read (u, num_threads, parameters), so a call carried between them bound a parameter matrix as a thread count. Bicop(family, rotation, parameters), from_family and parameters_to_tau keep theirs positional, where parameters specifies the model rather than the call. Keyword calls are unaffected (#345).

  • cop.pdf(u, pars) -> cop.pdf(u, parameters=pars)
  • cop.sample(n, False, [1, 2]) -> cop.sample(n, False, seeds=[1, 2])

• Name the observations u on Bicop / Vinecop's fit / select / from_data, which is what every evaluation method on the same class already called them. Renamed upstream rather than in the binding, whose docstrings are lifted verbatim (vinecopulib#781, #345).

  • cop.fit(data=u) -> cop.fit(u=u), and likewise on select / from_data

• Drop Vinecop.fit's num_threads argument, which duplicated FitControlsBicop.num_threads (#326).

  • vine.fit(u, controls, num_threads=4) -> vine.fit(u, FitControlsBicop(num_threads=4))

• Rename Kde1d's quantile to icdf, the name modern SciPy and torch.distributions use for the inverse distribution function, with no alias (#292).

• Rename Kde1d's edf to npars, the name Bicop and Vinecop already answer for the same quantity, with no alias. It was bound twice -- edf and n_parameters were the same getter -- so one class offered two live names for one number; npars is now the only spelling, on the compiled classes, on MarginBase and on the four contracts (#345).

  • kde.edf -> kde.npars
  • kde.quantile(p) -> kde.icdf(p)

• Serialize every margin the same way: to_json returns JSON text on SciPyMargin, TorchKde1d and TorchDistributionMargin as it already did on Bicop / Vinecop / Kde1d / Vinedist, and the reader is from_json(json) rather than from_json_payload(payload). A margin's own to_json builds its mapping and hands it to margin_json, which stamps kind and version and runs the shared codec -- so a non-finite parameter still travels as a tagged string and a subclass still reads back. register_margin_json takes a Callable[[str], Any] (#345).

... (truncated)

Changelog

Sourced from pyvinecopulib's changelog.

1.0.0 (2026-09-18)

This release ships the whole vine-modeling stack rather than the copula half of it. Vinedist and the new pyvinecopulib.margins layer put Sklar's theorem on the data scale; pyvinecopulib.torch runs the evaluation cascade on the GPU under autograd; pyvinecopulib.sklearn adds scikit-learn-compatible estimators; and the array-agnostic layer in pyvinecopulib.core hosts custom pair copulas and non-simplified vines. The public API is reorganized into the core, families, utils and margins subpackages, with the pre-reorganization top-level names kept as warning aliases until 2.0.

It also advances all three vendored C++ libraries, so nearly every tll and Kde1d fit moves. Read the migration guide in docs/migrating-to-1.0.md before upgrading from 0.7.x.

Breaking API changes in pyvinecopulib

• Use one argument order on every estimator: the observations first, then controls, then keyword-only whatever the object cannot infer. Vinecop.from_data took controls fifth, behind structure, so the call carried over from fit bound a controls object as a structure (#326).

  • pv.Vinecop.from_data(u, structure, matrix, var_types, controls) -> pv.Vinecop.from_data(u, controls, structure=..., matrix=..., var_types=...)
  • pv.Bicop.from_data(u, controls, var_types) -> pv.Bicop.from_data(u, controls, var_types=...)
  • pv.core.Kde1d.fit(x, weights) -> pv.core.Kde1d.fit(y, FitControlsKde1d(weights=...)), and likewise select; the kernel knobs and the weights are all FitControlsKde1d, and the observations are y, because x means exogenous covariates everywhere in the Python API (#339)
  • likewise on BicopBase, VinecopBase, VinedistBase and their torch subclasses; MarginBase already read this way

• Reorganize the public API into the core, families and utils subpackages; the family constants, Kde1d and the utility functions still resolve at the top level, but warn on access and are removed in 2.0 (#207, #292).

  • pyvinecopulib.gaussian -> pyvinecopulib.families.gaussian, and likewise every family constant and group
  • pyvinecopulib.Kde1d -> pyvinecopulib.core.Kde1d
  • pyvinecopulib.wdm -> pyvinecopulib.utils.wdm, and likewise sobol, ghalton, benchmark and pairs_copula_data

• Default selection_criterion to "aic" instead of "bic" on FitControlsBicop and FitControlsVinecop, matching the C++ and R defaults; pass selection_criterion="bic" to keep the previous selection (#251).

• Move every tll fit, and so every unqualified Bicop.from_data / Vinecop.from_data that selects one: the interpolation grid's margins are balanced across both sweep orders and iterated to convergence, and hfunc / hinv no longer floor the interpolated density at 1e-4 (#305, vinecopulib#751).

• Move every Kde1d fit: the plug-in bandwidth shifts by O(1/n), and a bounded fit shifts again under the new endpoint treatment (#312, kde1d#29).

• Fit a finite Kde1d bound with a local-linear boundary estimator by default; pass boundary_repair=False for the previous transformed fit (#312, kde1d#36).

• Read a discrete Kde1d's xmin / xmax as the integer support, so both the bounds and the data must be integers and the fitted grid runs half a unit wider at each end (#312, kde1d#37).

• Label a selected RVineStructure with the conditioned variable on the diagonal, so get_matrix() and order differ for the same model; densities and log-likelihoods do not (#251, vinecopulib#702).

• Rename the sampling entry points to sample, keeping the released spellings as warning aliases (#297).

  • Bicop.simulate(n) -> Bicop.sample(n), and likewise on Vinecop, RVineStructure and Kde1d
  • pyvinecopulib.simulate_uniform -> pyvinecopulib.utils.sample_uniform

• Return self from Bicop.fit / .select and Vinecop.fit / .select instead of None, so every estimator in the package composes the same way (#326).

• Make parameters, num_threads, seeds and randomize_discrete keyword-only on the evaluation methods of Bicop and Vinecop: each class was internally consistent and the two disagreed, Bicop reading (u, parameters, num_threads) where Vinecop read (u, num_threads, parameters), so a call carried between them bound a parameter matrix as a thread count. Bicop(family, rotation, parameters), from_family and parameters_to_tau keep theirs positional, where parameters specifies the model rather than the call. Keyword calls are unaffected (#345).

  • cop.pdf(u, pars) -> cop.pdf(u, parameters=pars)
  • cop.sample(n, False, [1, 2]) -> cop.sample(n, False, seeds=[1, 2])

• Name the observations u on Bicop / Vinecop's fit / select / from_data, which is what every evaluation method on the same class already called them. Renamed upstream rather than in the binding, whose docstrings are lifted verbatim (vinecopulib#781, #345).

  • cop.fit(data=u) -> cop.fit(u=u), and likewise on select / from_data

• Drop Vinecop.fit's num_threads argument, which duplicated FitControlsBicop.num_threads (#326).

  • vine.fit(u, controls, num_threads=4) -> vine.fit(u, FitControlsBicop(num_threads=4))

• Rename Kde1d's quantile to icdf, the name modern SciPy and torch.distributions use for the inverse distribution function, with no alias (#292).

• Rename Kde1d's edf to npars, the name Bicop and Vinecop already answer for the same quantity, with no alias. It was bound twice -- edf and n_parameters were the same getter -- so one class offered two live names for one number; npars is now the only spelling, on the compiled classes, on MarginBase and on the four contracts (#345).

  • kde.edf -> kde.npars
  • kde.quantile(p) -> kde.icdf(p)

• Serialize every margin the same way: to_json returns JSON text on SciPyMargin, TorchKde1d and TorchDistributionMargin as it already did on Bicop / Vinecop / Kde1d / Vinedist, and the reader is from_json(json) rather than from_json_payload(payload). A margin's own to_json builds its mapping and hands it to margin_json, which stamps kind and version and runs the shared codec -- so a non-finite parameter still travels as a tagged string and a subclass still reads back. register_margin_json takes a Callable[[str], Any] (#345).

  • margin.to_json() returns str, not dict

... (truncated)

Commits

• 1765dd0 ci: automate GitHub releases (#346)
• b526829 refactor(core)!: settle the four contracts, and delete the machinery they rep...
• 0a4a571 build: track the lockfile, and pin what was pinned everywhere but here (#337)
• 84f187a build(deps): bump the pins, and follow upstream downstream (#336)
• 27063ca feat(core): the extension surface a third-party subclass needs (#334)
• d01c6b6 chore: trim the comments that repeat themselves, and delete the dead ones (#331)
• b82de8f ci: bump pypa/cibuildwheel from 4.2.0 to 4.2.1 (#333)
• 4b88390 ci: bump vinecopulib/vinecopulib/.github/actions/install-dependencies from be...
• 261c56b feat(core)!: one extension-point architecture across all four layers (#326)
• e177e6e fix: close the Gate 1 and Gate 2 release audits (#320)
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.47%. Comparing base (ca9db44) to head (fd84094).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1949   +/-   ##
=======================================
  Coverage   87.47%   87.47%           
=======================================
  Files         430      430           
  Lines       11159    11159           
=======================================
  Hits         9761     9761           
  Misses       1398     1398           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.