fix(eval): handle None score in LocalEvalSampler to prevent TypeError crash

[nileshpatil6]

Problem

When a metric evaluation fails (transient API error, rate limiting, JSONDecodeError), local_eval_service.py catches the exception and returns an EvaluationResult with score=None. However, local_eval_sampler.py line 292 then calls round(eval_metric_result.score, 2) unconditionally, crashing with:

TypeError: type NoneType doesn't define __round__ method

Fixes #5403

Fix

Guard the round() call with a None check:

"score": round(eval_metric_result.score, 2) if eval_metric_result.score is not None else None,

[nileshpatil6]

I have read the Contributor License Agreement and I hereby agree to its terms.

[adk-bot]

Response from ADK Triaging Agent

Hello @nileshpatil6, thank you for submitting this pull request!

To help us review it more efficiently, could you please add a testing plan section to your PR description explaining how you tested these changes?

Additionally, since this is a bug fix, could you please provide logs or a screenshot demonstrating that the TypeError is resolved after your changes?

You can find more details in our contribution guidelines. Thanks!

[nileshpatil6]

Testing Plan

Manual verification:

from unittest.mock import MagicMock
from google.adk.optimization.local_eval_sampler import LocalEvalSampler

# Simulate a metric result with None score (happens when evaluation fails)
mock_metric_result = MagicMock()
mock_metric_result.score = None
mock_metric_result.metric_name = 'test_metric'
mock_metric_result.eval_status.name = 'NOT_EVALUATED'

mock_per_invocation = MagicMock()
mock_per_invocation.eval_metric_results = [mock_metric_result]

mock_eval_result = MagicMock()
mock_eval_result.eval_metric_result_per_invocation = [mock_per_invocation]

Before fix:

TypeError: type NoneType doesn't define __round__ method

at round(eval_metric_result.score, 2) when score is None.

After fix: The None score is passed through as-is (None) instead of crashing, consistent with how the upstream local_eval_service.py already uses None to represent a failed evaluation.

Reproducing the original crash (as described in issue #5403): trigger a transient API error or JSONDecodeError during metric evaluation via adk optimize — the sampler now handles it gracefully instead of crashing.

[rohityan]

Hi @nileshpatil6 , Thank you for your contribution! We appreciate you taking the time to submit this pull request. Please fix formatting errors by running autoformat.sh

[nileshpatil6]

Hi @nileshpatil6 , Thank you for your contribution! We appreciate you taking the time to submit this pull request. Please fix formatting errors by running autoformat.sh

Thank you for responding , I have Fixed the issue, Ran the formatting fix -- the ternary expression is now wrapped across multiple lines per pyink style. Pushed in fe19f31.