Skip to content

Improve reporting on robustness/ethical issues [GSK-1765] #1411

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

Merged
merged 3 commits into from
Oct 17, 2023
Merged

Improve reporting on robustness/ethical issues [GSK-1765] #1411

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f283990
Improve reporting on robustness/ethical issues
mattbit Sep 20, 2023
e43f487
Update phrasing in issue description
mattbit Oct 17, 2023
1dab597
Merge branch 'main' into task/GSK-1765
mattbit Oct 17, 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
32 changes: 24 additions & 8 deletions giskard/scanner/robustness/base_detector.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -134,26 +134,26 @@ def _detect_issues(
# Severity
issue_level = IssueLevel.MAJOR if fail_rate >= 2 * threshold else IssueLevel.MEDIUM

# Prepare examples dataframe
examples = original_data.df.loc[~passed, (feature,)].copy()
examples[f"{transformation_fn.name}({feature})"] = perturbed_data.df.loc[~passed, feature]
examples["Original prediction"] = original_pred.prediction[~passed]
examples["Prediction after perturbation"] = perturbed_pred.prediction[~passed]

# Description
desc = "When we perturb the content of feature “{feature}” with the transformation “{transformation_fn}” (see examples below), your model changes its prediction in about {fail_rate_percent}% of the cases. We expected the predictions not to be affected by this transformation."

failed_size = (~passed).sum()
slice_size = len(passed)

issue = Issue(
model,
dataset,
group=self._issue_group,
level=issue_level,
transformation_fn=transformation_fn,
description=desc,
features=[feature],
meta={
"feature": feature,
"domain": f"Feature `{feature}`",
"deviation": f"{round(fail_rate * 100, 2)}% of tested samples changed prediction after perturbation",
"deviation": f"{failed_size}/{slice_size} tested samples ({round(fail_rate * 100, 2)}%) changed prediction after perturbation",
"failed_size": failed_size,
"slice_size": slice_size,
"fail_rate": fail_rate,
"fail_rate_percent": round(fail_rate * 100, 2),
"metric": "Fail rate",
Expand All @@ -164,10 +164,26 @@ def _detect_issues(
"perturbed_data_slice_predictions": perturbed_pred,
},
importance=fail_rate,
examples=examples,
tests=_generate_robustness_tests,
)

# Add examples
examples = original_data.df.loc[~passed, (feature,)].copy()
examples[f"{transformation_fn.name}({feature})"] = perturbed_data.df.loc[~passed, feature]

examples["Original prediction"] = original_pred.prediction[~passed]
examples["Prediction after perturbation"] = perturbed_pred.prediction[~passed]

if model.is_classification:
examples["Original prediction"] = examples["Original prediction"].astype(str)
examples["Prediction after perturbation"] = examples["Prediction after perturbation"].astype(str)
ps_before = pd.Series(original_pred.probabilities[~passed], index=examples.index)
ps_after = pd.Series(perturbed_pred.probabilities[~passed], index=examples.index)
examples["Original prediction"] += ps_before.apply(lambda p: f" (p = {p:.2f})")
examples["Prediction after perturbation"] += ps_after.apply(lambda p: f" (p = {p:.2f})")

issue.add_examples(examples)

issues.append(issue)

return issues
Expand Down