Expose metric results of Step Efficiency

[AmaliMatharaarachchi]

Description

Document is incomplete and vague and does not show how standalone runs can be performed for trace based matrics. See section https://deepeval.com/docs/metrics-step-efficiency#as-a-standalone. give working sample for convo_test_case

evals_iterator() currently runs trace-based metrics (e.g., StepEfficiency, TaskCompletion) internally and uploads them to a ConfidentAI test run, but there is no way to access these metric results from Python code.

This prevents users from using trace-based metrics inside custom evaluation workflows.

Expected behaviour

print(metric.score) # at the end, expected metric results should be printable

[penguine-ip]

Hey @AmaliMatharaarachchi thanks for bringing this up. @A-Vamshi will the verbose logs give the info on the results?

[A-Vamshi]

Hey @AmaliMatharaarachchi,
Apologies for the late reply on this one, the docs are updated now, however, to answer your question you can use the metric scores right after running the evals_iterator as following:

metric = StepEfficiencyMetric()

for golden in dataset.evals_iterator(metrics=[metric]):
    agent(golden.input)

print(metric.score, metric.reason)

The agentic metrics require the full trace of the application so it's not usable as a standalone where you do metric.measure(test_case) but you can use the scores after running in the evals_iterator or observe tag.

Please let me know if you have any doubts or any feature requests on these metrics.

And as @penguine-ip mentioned, you can see the scores and reasons in verbose mode as well.