When I manually call log_evaluations, no data shown in frontend

[qlibp]

Is it necessary to make LLM API call before calling log_evaluations?

[dosubot[bot]]

Hey @qlibp! I'm here to help you with any bugs, questions, or contributions you have. Let me know how I can assist you!

It is not necessary to make an LLM API call before using log_evaluations. However, there are specific prerequisites and steps you need to follow to ensure that the data is displayed in the frontend:

1. Environment Variables: Make sure the environment variables PHOENIX_CLIENT_HEADERS and PHOENIX_COLLECTOR_ENDPOINT are set correctly.

2. DataFrame Structure: Ensure your DataFrame is structured correctly:

   • For span evaluations, it should include span_id as an index or column.
   • For document evaluations, it should include both span_id and document_position as indices or columns.

3. Evaluation Name: Provide an evaluation name using the eval_name parameter.

4. DataFrame Columns: The DataFrame should have values under the score, label, or explanation columns.

5. Deprecation Notice: The log_evaluations function is deprecated. Use px.Client().log_evaluations(*evaluations) instead.

Ensure these conditions are met, and your data should appear in the frontend ^[1][2].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other}

[qlibp]

not work

[axiomofjoy]

Hey @qlibp, it's not necessary, but very often, the input to log_evaluations will come from run_evals and LLM output.

https://docs.arize.com/phoenix/tracing/how-to-interact-with-traces/llm-evaluations#document-evaluations

[axiomofjoy]

Filed an issue here