Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support inference with RenyiELBO for local latent variable models #3123

Open
wants to merge 7 commits into
base: dev
Choose a base branch
from

Conversation

OlaRonning
Copy link
Member

@OlaRonning OlaRonning commented Aug 2, 2022

Addresses #3104

TODO

  • Add Local latent variable model test case
  • Correct RenyiELBO logsumexp reduction dim for LLV model

@OlaRonning OlaRonning added the WIP label Aug 2, 2022
@OlaRonning
Copy link
Member Author

Should I ignore notebooks when using black? When running black[jupyter]==22.6.0, I get 48 files changed.

@fritzo
Copy link
Member

fritzo commented Aug 9, 2022

Yes could you please ignore notebooks? The easiest way to do so is to run black through make format. Alternatively we could set up a pyproject.toml file with the black exclusion list

@OlaRonning
Copy link
Member Author

Yes could you please ignore notebooks?

Sure, no problem.

The easiest way to do so is to run black through make format. Alternatively we could set up a pyproject.toml file with the black exclusion list.

Looking at the format target in the Makefile, it seems it's only black that needs a section in the TOML file. I'll configure a pyproject.toml and update the Makefile.



@pytest.mark.stage("integration", "integration_batch_2")
class OneWayNormalRandomEffects(TestCase):
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

how much time do these tests take?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

how much time do these tests take?

In CI :

Time [sec]
test_renyi_nonreparameterized 50.67
test_renyi_reparameterized 36.59
test_renyi_vectorized 36.32

So all are in the top ten slowest for batch 2.

@OlaRonning
Copy link
Member Author

@martinjankowiak, can you suggest a plating structure where we need to compute the dependence (i.e., all_dims - sum_dims != num_particles_vectorized)? I've been playing around with the current implementation but cannot come up with one.

Copy link
Collaborator

@martinjankowiak martinjankowiak left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@OlaRonning will this still give results for a local latent variable model like a VAE with a data plate? why doesn't this logic sum out the data plate dimension?

@OlaRonning
Copy link
Member Author

@martinjankowiak, sorry for the late reply.

why doesn't this logic sum out the data plate dimension?

From my understanding, this corresponds to including the data plate dimension in sum_dims, or have I misunderstood something?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants