Show Metadata for Each Example in Experiment View

Zhuyi X.

·May 02, 2025 11:39 PM

is there a way to show some metadata for each example in the experiment view, e.g.

11 comments

· Sorted by Oldest

RunLLM
APP
·
Yes, Phoenix supports displaying metadata for each example in the experiment view. When you upload a dataset, you can specify metadata fields, and these are preserved and accessible in the experiment results, including in the UI and when exporting as a DataFrame. The experiment view allows you to see full metadata for each example, as recent updates have added the ability to view experiment metadata in full detail (<https://github.com/Arize-ai/phoenix/blob/main/CHANGELOG.md|changelog>, <https://github.com/Arize-ai/phoenix/pull/3686|PR #3686>). To include metadata, ensure your dataset upload includes a metadata column, and Phoenix will propagate this to experiments. You can access this metadata in the UI or programmatically via the `.as_dataframe()` method on experiment or dataset objects (<https://docs.arize.com/phoenix/datasets-and-experiments/concepts-datasets|docs>, <https://github.com/Arize-ai/phoenix/blob/main/tutorials/experiments/llama-index/pairwise_eval.ipynb|example>). Would you like more detail on how to upload or view metadata? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/CHANGELOG.md|Phoenix CHANGELOG.md> - <https://github.com/Arize-ai/phoenix/pull/3686|PR: add the ability to view experiment metadata in full> - <https://github.com/Arize-ai/phoenix/blob/main/tutorials/experiments/llama-index/pairwise_eval.ipynb|Llama Index Pairwise Eval Example>
💡 Hint: Mention RunLLM in the thread for followups.
Xander S.
·
Hey Zhuyi X., are you interested in attaching metadata to the examples, or to the runs in the experiments?
Zhuyi X.
·
to the examples, e.g. each example could have an ID, which can be used to inspect it more closely in other tools
Zhuyi X.
·
the View metadata in the triple dot refers to the metadata of the experiment run, right?
Zhuyi X.
·
The motivation is like for an exmple in an experiment, if the output doesn't match the reference, i'd like to understand why by looking at more info about the example.
👍1
Xander S.
·
That metadata is the metadata associated with the experiment run, I believe. Each example has an editable metadata field. It doesn't look like we currently have affordances to view that metadata as part of the experiment run results, though.
💡1
Zhuyi X.
·
then in such case,
for an exmple in an experiment, if the output doesn't match the reference, i'd like to understand
how to locate the ID of the example quickly?
Zhuyi X.
·
oh i see, I can expand the input to see the metadata
Zhuyi X.
·
👍1
Xander S.
·
Your use-case definitely makes sense. Does that provide what you needed?
Zhuyi X.
·
Does that provide what you needed?
yeah, I see so. Thanks!
👍1

RunLLM
APP
·
Yes, Phoenix supports displaying metadata for each example in the experiment view. When you upload a dataset, you can specify metadata fields, and these are preserved and accessible in the experiment results, including in the UI and when exporting as a DataFrame. The experiment view allows you to see full metadata for each example, as recent updates have added the ability to view experiment metadata in full detail (<https://github.com/Arize-ai/phoenix/blob/main/CHANGELOG.md|changelog>, <https://github.com/Arize-ai/phoenix/pull/3686|PR #3686>). To include metadata, ensure your dataset upload includes a metadata column, and Phoenix will propagate this to experiments. You can access this metadata in the UI or programmatically via the `.as_dataframe()` method on experiment or dataset objects (<https://docs.arize.com/phoenix/datasets-and-experiments/concepts-datasets|docs>, <https://github.com/Arize-ai/phoenix/blob/main/tutorials/experiments/llama-index/pairwise_eval.ipynb|example>). Would you like more detail on how to upload or view metadata? Sources: - <https://github.com/Arize-ai/phoenix/blob/main/CHANGELOG.md|Phoenix CHANGELOG.md> - <https://github.com/Arize-ai/phoenix/pull/3686|PR: add the ability to view experiment metadata in full> - <https://github.com/Arize-ai/phoenix/blob/main/tutorials/experiments/llama-index/pairwise_eval.ipynb|Llama Index Pairwise Eval Example>
💡 Hint: Mention RunLLM in the thread for followups.
Xander S.
·
Hey Zhuyi X., are you interested in attaching metadata to the examples, or to the runs in the experiments?
Zhuyi X.
·
to the examples, e.g. each example could have an ID, which can be used to inspect it more closely in other tools
Zhuyi X.
·
the View metadata in the triple dot refers to the metadata of the experiment run, right?
Zhuyi X.
·
The motivation is like for an exmple in an experiment, if the output doesn't match the reference, i'd like to understand why by looking at more info about the example.
👍1
Xander S.
·
That metadata is the metadata associated with the experiment run, I believe. Each example has an editable metadata field. It doesn't look like we currently have affordances to view that metadata as part of the experiment run results, though.
💡1
Zhuyi X.
·
then in such case,
for an exmple in an experiment, if the output doesn't match the reference, i'd like to understand
how to locate the ID of the example quickly?
Zhuyi X.
·
oh i see, I can expand the input to see the metadata
Zhuyi X.
·
👍1
Xander S.
·
Your use-case definitely makes sense. Does that provide what you needed?
Zhuyi X.
·
Does that provide what you needed?
yeah, I see so. Thanks!
👍1