Refactor funsor.distributions #319
Closed
Conversation
This file contains 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
eb8680
commented
Mar 3, 2020
Comment on lines
+59
to
+67
| def _infer_value_shape(cls, **kwargs): | ||
| # rely on the underlying distribution's logic to infer the event_shape | ||
| instance = cls.dist_class(**{k: _dummy_tensor(v.output) for k, v in kwargs.items()}) | ||
| out_shape = instance.event_shape | ||
| if isinstance(instance.support, torch.distributions.constraints._IntegerInterval): | ||
| out_dtype = instance.support.upper_bound + 1 | ||
| else: | ||
| out_dtype = 'real' | ||
| return Domain(dtype=out_dtype, shape=out_shape) |
There was a problem hiding this comment.
This method is the main innovation in this PR: it turns out we can reuse the distributions' existing event_shape and support inference logic to generically infer the shape of the fresh value input, rather than reimplementing that logic by hand for every distribution we add.
eb8680
commented
Mar 3, 2020
| return dist_class | ||
|
|
||
|
|
||
| class BernoulliProbs(dist.Bernoulli): |
There was a problem hiding this comment.
I added a few of these wrappers for distributions with multiple parametrizations, but these should probably be moved upstream to Pyro or PyTorch.
|
Closing in favor of #320 which is similar but narrower in scope |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
This PR adds a replacement for the
funsor.distributionsmodule in which PyTorch distributions are wrapped programmatically andto_funsorandto_dataare defined generically. The main API change in this version is that distributions now take anameargument rather than avalue, making them behave a bit more likeGaussianorDelta, but I'm open to reverting that.Remaining tasks:
tensor_to_funsorso that distribution shape inference workseagerrules for distribution evaluation to the new moduledistributionscode