Forward Sampling

sample_corpus draws length-fixed yields from the chart's length-conditional distribution \(p(s \mid \text{length} = L, \mathbf{w}) = Z(s; \mathbf{w}) / \sum_{s' \text{ of length } L} Z(s'; \mathbf{w})\). The implementation enumerates every length-\(L\) sequence over the deduction's surface vocabulary, evaluates \(\log Z\) exactly via the chart, softmaxes the log-weights, and draws a multinomial. The procedure is exact (no MCMC over derivations); the \(|V|^L\) enumeration is the cost of this implementation.

sample

Sample fixed-length yields from a weighted deduction system.

sample_corpus draws length-fixed token sequences from the length-conditional distribution induced by the chart's weights:

.. math::

p(s \mid \text{length} = L,\, \mathbf{w})
\;=\; \frac{Z(s; \mathbf{w})}
             {\sum_{s' \text{ of length } L} Z(s'; \mathbf{w})}.

The sampler enumerates all :math:|V|^L sequences, evaluates each chart partition score, normalizes those scores, and draws from the resulting distribution. The chart marginalizes the derivation forest for each sequence.

sample_corpus

sample_corpus(ded: DeductionSystem, *, length: int, n_samples: int, seed: int | None = None) -> list[list[str]]

Sample n_samples yields of length length from the chart's length-conditional distribution under the deduction's current parameters.

PARAMETER	DESCRIPTION
ded	The deduction with materialised parameters. TYPE: DeductionSystem
length	Length of yields to enumerate. TYPE: int
n_samples	Number of sentences to draw. TYPE: int
seed	Seed for the multinomial draws. TYPE: int DEFAULT: None

RETURNS	DESCRIPTION
list of list of str	n_samples yields, each a list of length words, drawn with replacement in proportion to their goal weights.

RAISES	DESCRIPTION
ValueError	If the deduction's vocabulary cannot be determined, or if no yield of length length parses.

Source code in src/quivers/stochastic/deduction/sample.py

def sample_corpus(
    ded: DeductionSystem,
    *,
    length: int,
    n_samples: int,
    seed: int | None = None,
) -> list[list[str]]:
    """Sample ``n_samples`` yields of length ``length`` from the
    chart's length-conditional distribution under the deduction's
    current parameters.

    Parameters
    ----------
    ded : DeductionSystem
        The deduction with materialised parameters.
    length : int
        Length of yields to enumerate.
    n_samples : int
        Number of sentences to draw.
    seed : int, optional
        Seed for the multinomial draws.

    Returns
    -------
    list of list of str
        ``n_samples`` yields, each a list of ``length`` words, drawn
        with replacement in proportion to their goal weights.

    Raises
    ------
    ValueError
        If the deduction's vocabulary cannot be determined, or if no
        yield of length ``length`` parses.
    """
    vocab = _vocabulary(ded)
    if not vocab:
        raise ValueError(
            "sample_corpus: cannot determine the deduction's "
            "vocabulary; set ``ded._vocabulary`` explicitly or "
            "call ``materialise_parameters`` first"
        )

    yields: list[list[str]] = []
    log_weights: list[torch.Tensor] = []
    for combo in itertools.product(vocab, repeat=length):
        chart = ded(list(combo))
        w = chart.goal_weight()
        if torch.isfinite(w):
            yields.append(list(combo))
            log_weights.append(w)
    if not yields:
        raise ValueError(
            f"sample_corpus: no yield of length {length} parses under "
            f"the deduction's current parameters"
        )
    logw = torch.stack([w.detach() for w in log_weights])
    probs = torch.softmax(logw, dim=0)
    gen = torch.Generator()
    if seed is not None:
        gen.manual_seed(seed)
    idxs = torch.multinomial(probs, n_samples, replacement=True, generator=gen)
    return [yields[int(i.item())] for i in idxs]