Families/ adding base version of exponential family#67
Open
domosedy wants to merge 12 commits into
Open
Conversation
LeonidElkin
reviewed
Feb 20, 2026
Desiment
requested changes
Feb 21, 2026
domosedy
force-pushed
the
exponential-family
branch
2 times, most recently
from
779438f to
12516ff
Compare
LeonidElkin
requested changes
May 1, 2026
domosedy
force-pushed
the
exponential-family
branch
2 times, most recently
from
f67f3fd to
2d470c9
Compare
domosedy
force-pushed
the
exponential-family
branch
from
2d470c9 to
189afea
Compare
domosedy
force-pushed
the
exponential-family
branch
3 times, most recently
from
1605771 to
1d58077
Compare
domosedy
force-pushed
the
exponential-family
branch
from
83f5584 to
2500678
Compare
LeonidElkin
reviewed
May 25, 2026
domosedy
force-pushed
the
exponential-family
branch
from
ec832d5 to
d5a2be1
Compare
domosedy
force-pushed
the
exponential-family
branch
from
d5a2be1 to
804cfa4
Compare
LeonidElkin
reviewed
May 31, 2026
… for exponential family
LeonidElkin
requested changes
Jun 9, 2026
Comment on lines
+328
to
+339
| return ContinuousExponentialClassFamily( | ||
| log_partition=conjugate_log_partition, | ||
| sufficient_statistics=conjugate_sufficient, | ||
| normalization_constant=lambda _: 1, | ||
| support=self._parameter_space, | ||
| sufficient_statistics_values=self._parameter_space, | ||
| parameter_space=PredicateSupport(predicate=conjugate_sufficient_accepts), | ||
| name=self.name, | ||
| distr_type=self._distr_type, | ||
| distr_parametrizations=self.parametrization_names, | ||
| support_by_parametrization=self.support_resolver, | ||
| ) |
Collaborator
There was a problem hiding this comment.
_support is changed for the conjugate/transformed family, but the family still reuses self.support_resolver.
This makes the density use the new support while distribution.support can remain None or point to the original support. Expose the same transformed support through support_by_parametrization as well, e.g. for transform() return a resolver for PredicateSupport(new_support) and for the conjugate prior return the natural-parameter support.
Comment on lines
+385
to
+396
| return ContinuousExponentialClassFamily( | ||
| log_partition=self._log_partition, | ||
| sufficient_statistics=new_sufficient, | ||
| normalization_constant=new_normalization, | ||
| support=PredicateSupport(predicate=new_support), | ||
| parameter_space=self._parameter_space, | ||
| sufficient_statistics_values=self._sufficient_statistics_values, | ||
| name=f"Transformed{self._name}", | ||
| distr_type=self._distr_type, | ||
| distr_parametrizations=self.parametrization_names, | ||
| support_by_parametrization=self.support_resolver, | ||
| ) |
Collaborator
There was a problem hiding this comment.
Same issue as in comment above
Comment on lines
+446
to
+489
| def posterior_hyperparameters( | ||
| self, | ||
| parametrization: ExponentialConjugateHyperparameters, | ||
| sample: list[Any] | Any, | ||
| ) -> ExponentialConjugateHyperparameters: | ||
| """ | ||
| Update the conjugate prior hyperparameters given observed data. | ||
|
|
||
| For prior hyperparameters ``(nu_0, n_0)`` and observations ``x_i``, | ||
| the posterior hyperparameters are | ||
|
|
||
| ``nu = nu_0 + sum_i T(x_i)`` and ``n = n_0 + N``. | ||
|
|
||
| Parameters | ||
| ---------- | ||
| parametrization : ExponentialConjugateHyperparameters | ||
| Current conjugate hyperparameters. | ||
| sample : list[Any] or Any | ||
| Observations used for the update. A non-string iterable is treated | ||
| as a sample; any other value is treated as one observation. | ||
|
|
||
| Returns | ||
| ------- | ||
| ExponentialConjugateHyperparameters | ||
| Updated hyperparameters after incorporating ``sample``. | ||
| """ | ||
| if hasattr(sample, "__iter__") and not isinstance(sample, str): | ||
| posterior_effective_suff_stat_value = np.array( | ||
| parametrization.effective_suff_stat_value | ||
| ) + np.sum( | ||
| [self._sufficient(x) for x in sample], | ||
| axis=0, | ||
| ) | ||
| posterior_effective_sample_size = parametrization.effective_sample_size + len(sample) | ||
| else: | ||
| posterior_effective_suff_stat_value = np.array( | ||
| parametrization.effective_suff_stat_value, | ||
| ) + np.asarray(self._sufficient(sample)) # type: ignore[arg-type] | ||
| posterior_effective_sample_size = parametrization.effective_sample_size + 1 | ||
|
|
||
| return ExponentialConjugateHyperparameters( | ||
| effective_suff_stat_value=posterior_effective_suff_stat_value, | ||
| effective_sample_size=posterior_effective_sample_size, | ||
| ) |
Collaborator
There was a problem hiding this comment.
AI finding
hasattr(sample, "__iter__") misclassifies zero-dimensional NumPy arrays as samples.
posterior_hyperparameters(..., sample=np.asarray(0.5)) enters the iterable branch and raises TypeError: iteration over a 0-d array. Normalize the input explicitly, e.g. distinguish scalar/0-D observations from batches with np.ndim(sample).
Comment on lines
+511
to
+528
| def conjugate_log_partition( | ||
| parametrization: ExponentialConjugateHyperparameters, | ||
| ) -> Number: | ||
| conjugate_value = self.conjugate_prior_family._log_partition( | ||
| parametrization.transform_to_base_parametrization().theta | ||
| ) | ||
| return cast(Number, np.exp(conjugate_value).item()) | ||
|
|
||
| def posterior_density(parametrization: Parametrization, x: NumericArray) -> Number: | ||
| parametrization = cast(ExponentialConjugateHyperparameters, parametrization) | ||
| if not self._support.contains(np.asarray(x)): | ||
| return cast(np.float32, 0.0) | ||
| return cast( | ||
| np.float32, | ||
| self._normalization(x) | ||
| * conjugate_log_partition(parametrization) | ||
| / conjugate_log_partition( | ||
| self.posterior_hyperparameters(parametrization=parametrization, sample=[x]) |
Collaborator
There was a problem hiding this comment.
AI finding
posterior_density recomputes the conjugate-prior family and its normalizing integral on every density evaluation.
conjugate_log_partition(...) accesses self.conjugate_prior_family, and each call eventually runs nquad; posterior_density calls it twice per x. Construct/cache the conjugate family once in posterior_predictive and avoid recomputing invariant normalizers for every point.
Comment on lines
+182
to
+189
| family_characteristics: CharacteristicsMap = { | ||
| CharacteristicName.PDF: self.density, | ||
| CharacteristicName.MEAN: self._mean, | ||
| CharacteristicName.VAR: self._var, | ||
| } | ||
| distr_characteristics = dict(distr_characteristics or {}) | ||
| merged_characteristics = dict(family_characteristics) | ||
| merged_characteristics.update(distr_characteristics) |
Collaborator
There was a problem hiding this comment.
The constructor now preserves user characteristics, but the merge is still shallow and does not keep the generic exponential-family methods under separate labels.
If a user provides MEAN/VAR/PDF, the generic fallback is replaced entirely. The intended contract is to "expand under other labels". Merge by label and add a test for the collision case, not only for adding CDF.
… for exponential family
domosedy
force-pushed
the
exponential-family
branch
from
8b99bff to
16115cb
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.