VinecopLike

class VinecopLike(*args, **kwargs)

Contract for a post-fit vine-copula evaluator.

A vine evaluator exposes the joint pdf / cdf, the rosenblatt and inverse_rosenblatt transforms, and a simulate sampler, plus a structure attribute (the RVineStructure it was built on). The optional keyword-only x covariate matrix (row-aligned with u) is the conditional extension — left None for the usual (unconditional) case. Subclass VinecopBase to get every method from a small set of hooks; pyvinecopulib.core.Vinecop and pyvinecopulib.torch.TorchVinecop are the reference implementations.

Methods

`cdf`(u, *[, x, N, num_threads, seeds])	Joint vine-copula distribution `C(u)` via Monte-Carlo.
`inverse_rosenblatt`(u, *[, x, num_threads])	Inverse Rosenblatt transform: independent uniforms to dependent uniforms.
`pdf`(u, *[, x, num_threads])	Joint vine-copula density `c(u_1, ..., u_d)` at each observation.
`rosenblatt`(u, *[, x, num_threads])	Rosenblatt transform: dependent uniforms to independent uniforms.
`simulate`(n, *[, x, qrng, num_threads, seeds])	Draw `n` samples from the fitted vine copula.

See also

pyvinecopulib.core.VinecopBase: Canonical partial implementation to subclass.
pyvinecopulib.core.Vinecop: The reference vine.
BicopLike: The pair-copula contract.

Examples

Host copulas in a vine by subclassing VinecopBase; the only required hook is _get_pair_copula. Under the default SimplifiedContext it is a classic (unconditional, simplified) vine, and hosting Bicop pairs reproduces from_structure():

import numpy as np
import pyvinecopulib as pv
from pyvinecopulib.core import VinecopBase

class ListVinecop(VinecopBase[np.ndarray]):
  def __init__(self, pairs, structure):
    self._pairs = pairs
    self._bind_vine(structure)          # SimplifiedContext by default

  def _get_pair_copula(self, tree, edge):
    return self._pairs[tree][edge]

struct = pv.RVineStructure.from_order([1, 2, 3])
g = pv.families.gaussian
pairs = [
  [pv.Bicop(family=g, parameters=np.array([[0.5]])),
   pv.Bicop(family=g, parameters=np.array([[0.4]]))],
  [pv.Bicop(family=g, parameters=np.array([[0.2]]))],
]
vine = ListVinecop(pairs, struct)
ref = pv.Vinecop.from_structure(structure=struct, pair_copulas=pairs)
u = np.random.default_rng(0).uniform(size=(4, 3))
np.allclose(vine.pdf(u), ref.pdf(u))    # -> True

Attributes

structure

Methods

`__init__`
`cdf`	Joint vine-copula distribution `C(u)` via Monte-Carlo.
`inverse_rosenblatt`	Inverse Rosenblatt transform: independent uniforms to dependent uniforms.
`pdf`	Joint vine-copula density `c(u_1, ..., u_d)` at each observation.
`rosenblatt`	Rosenblatt transform: dependent uniforms to independent uniforms.
`simulate`	Draw `n` samples from the fitted vine copula.