Get explanation along with scores

[jkhetan]

The evaluate call on a data set and list of metrics returns only the scores.

Isn't there any way to also get a corresponding explanation? Especially if the scores are low and I have hundreds of questions, it is almost impossible to rationalize them manually.

Any help would be greatly appreciated.

[anistark]

Continuing from #2296

I guess you're trying to access .get('reason') directly on the prompt output. However, faithfulness metric's prompt output (NLIStatementOutput) is a Pydantic model with a statements field (a list). The reasoning is nested one level deeper: statements[i].reason, not at the top level.

Try to access something like results.traces[i][metric_name][prompt_name]['output']