feat: Serializable execution details (#3358)

* Simplify output types for `llm_classify` with execution info * Update notebook * Ruff 🐶 * Fix typos
Arize-ai · May 31, 2024 · fc74513 · fc74513
1 parent dc18123
commit fc74513
Show file tree

Hide file tree

Showing 3 changed files with 1,656 additions and 10 deletions.
diff --git a/packages/phoenix-evals/src/phoenix/evals/classify.py b/packages/phoenix-evals/src/phoenix/evals/classify.py
@@ -249,8 +249,16 @@ def _run_llm_classification_sync(input_data: pd.Series[Any]) -> ParsedLLMRespons
             **({"explanation": explanations} if provide_explanation else {}),
             **({"prompt": prompts} if include_prompt else {}),
             **({"response": responses} if include_response else {}),
-            **({"exceptions": all_exceptions} if include_exceptions else {}),
-            **({"execution_status": classification_statuses} if include_exceptions else {}),
+            **(
+                {"exceptions": [[repr(exc) for exc in excs] for excs in all_exceptions]}
+                if include_exceptions
+                else {}
+            ),
+            **(
+                {"execution_status": [status.value for status in classification_statuses]}
+                if include_exceptions
+                else {}
+            ),
         },
         index=dataframe.index,
     )

diff --git a/packages/phoenix-evals/tests/phoenix/evals/functions/test_classify.py b/packages/phoenix-evals/tests/phoenix/evals/functions/test_classify.py
@@ -477,10 +477,10 @@ def test_classify_exits_on_missing_input(
     )
 
     assert classification_df["execution_status"].tolist() == [
-        ClassificationStatus.COMPLETED,
-        ClassificationStatus.COMPLETED,
-        ClassificationStatus.MISSING_INPUT,
-        ClassificationStatus.DID_NOT_RUN,
+        ClassificationStatus.COMPLETED.value,
+        ClassificationStatus.COMPLETED.value,
+        ClassificationStatus.MISSING_INPUT.value,
+        ClassificationStatus.DID_NOT_RUN.value,
     ]
 
     exceptions = classification_df["exceptions"].tolist()
@@ -525,10 +525,10 @@ def test_classify_skips_missing_input_with_when_exit_on_error_false(
     )
 
     assert classification_df["execution_status"].tolist() == [
-        ClassificationStatus.COMPLETED,
-        ClassificationStatus.COMPLETED,
-        ClassificationStatus.MISSING_INPUT,
-        ClassificationStatus.FAILED,
+        ClassificationStatus.COMPLETED.value,
+        ClassificationStatus.COMPLETED.value,
+        ClassificationStatus.MISSING_INPUT.value,
+        ClassificationStatus.FAILED.value,
     ]
 
     exceptions = classification_df["exceptions"].tolist()