Skip to content

Fix/use score for moment estimate #3

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

Open
dongyoung-go wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix/use score for moment estimate #3

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c624355
use score for feature moment estimate
dongyoung-go Sep 4, 2023
eaa9191
Use score fn for pipeline
dongyoung-go Sep 4, 2023
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
8 changes: 4 additions & 4 deletions disco/scorers/pipeline_scorer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,8 +26,8 @@ def __init__(self, label, params, temperature=1.0):
self.pipeline = pipeline(**params)
self.temperature = temperature

def log_score(self, samples, _):
"""computes the log-scores of the samples
def score(self, samples, _):
"""computes the scores of the samples
from the label returned by the pipeline

Parameters
Expand All @@ -37,9 +37,9 @@ def log_score(self, samples, _):

Returns
-------
tensor of log-probabilities"""
tensor of scores"""

return torch.log(
return (
torch.tensor(
[[r_i["score"] for r_i in r if self.label == r_i["label"]][0]
for r in self.pipeline([s.text for s in samples], return_all_scores=True)]
Expand Down
2 changes: 1 addition & 1 deletion disco/tuners/tuner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -187,7 +187,7 @@ def _update_features_moments(self, context, samples, proposal_log_scores, model_
logweights = model_log_scores - proposal_log_scores
importance_ratios = torch.exp(logweights)
for (label, feature) in self.features:
proposal_moment_pointwise_estimates = feature.log_score(samples, context).exp().to(device)
proposal_moment_pointwise_estimates = feature.score(samples, context).to(device)
self.features_moments_proposal[label][context] += proposal_moment_pointwise_estimates
self.features_moments_target[label][context] += importance_ratios * proposal_moment_pointwise_estimates

Expand Down