Implement a detach() helper that works on distributions #2599
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
fritzo
commented
Aug 13, 2020
|
|
||
| [isort] | ||
| line_length = 120 | ||
| not_skip = __init__.py |
There was a problem hiding this comment.
fixes a new error emitted by isort 5.0
fritzo
commented
Aug 13, 2020
| @@ -1,11 +1,10 @@ | |||
| [flake8] | |||
| max-line-length = 120 | |||
| exclude = docs/src, build, dist, .ipynb_checkpoints | |||
| extend-ignore = E741 | |||
There was a problem hiding this comment.
ignore errors about test assertions assert type(x) is type(y)
eb8680
reviewed
Aug 13, 2020
No, the |
|
OK, I've added a low-level jit test to check that |
1 task
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Outdated suggestions cannot be applied.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.
Addresses #2598, pytorch/pytorch#25783
This implements a function
detach(-)that creates a deep copy of an arbitrary Python object but with all tensors detached (and without copying underlying tensor data). Then main use case is in inference algorithms that require mixed detachment of gradients, e.g.d = MyDist(param)followed byd.rsample()butdetach(d).log_prob(x).This works by overriding
deepcopy's behavior to use a modified memoizer dict. This seems like the cleanest way to apply arbitrary transforms on leaves of python objects (similar to JAX's PyTree). We could in principle use this trick for other leaf transforms.Tested
TraceGraph_ELBOto use the new helper; should be covered by existing tests.cc @iffsid