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.

Sign up for GitHub

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

(WIP?) vectorized log_likelihood function for NumPyro #2390

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

(WIP?) vectorized log_likelihood function for NumPyro #2390

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 10 additions & 4 deletions arviz/data/io_numpyro.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -181,20 +181,26 @@ def sample_stats_to_xarray(self):
@requires("posterior")
@requires("model")
def log_likelihood_to_xarray(self):
"""Extract log likelihood from NumPyro posterior."""
"""Extract log likelihood from NumPyro posterior using vectorization."""
if not self.log_likelihood:
return None
Copy link
Member

Choose a reason for hiding this comment

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

Suggested change
return None
return None
import jax


data = {}
if self.observations is not None:
samples = self.posterior.get_samples(group_by_chain=False)
if hasattr(samples, "_asdict"):
samples = samples._asdict()
log_likelihood_dict = self.numpyro.infer.log_likelihood(
self.model, samples, *self._args, **self._kwargs
)

# Vectorized log likelihood calculation using jax.vmap
log_likelihood_dict = jax.vmap(lambda single_sample:
self.numpyro.infer.log_likelihood(self.model, single_sample, *self._args, **self._kwargs)
Copy link
Member

Choose a reason for hiding this comment

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

Suggested change
self.numpyro.infer.log_likelihood(self.model, single_sample, *self._args, **self._kwargs)
self.numpyro.infer.log_likelihood(self.model, single_sample, *self._args, batch_ndims=0, **self._kwargs)

It doesn't work without this because batching is not taken care of directly in vmap but this function expects a batch dimension too and fails when it is not there (or seemingly changes with the different variables)

)(samples)

# Process the log likelihood results
for obs_name, log_like in log_likelihood_dict.items():
shape = (self.nchains, self.ndraws) + log_like.shape[1:]
data[obs_name] = np.reshape(np.asarray(log_like), shape)

return dict_to_dataset(
data,
library=self.numpyro,
Expand Down
Loading