chore: import upstream snapshot with attribution

examples/flows/evaluation/eval-classification-accuracy-maf/aggregation.py

@@ -0,0 +1,6 @@
+from typing import Dict, List
+
+
+def aggregate(grades: List[str]) -> Dict[str, float]:
+    accuracy = round((grades.count("Correct") / len(grades)), 2) if grades else 0.0
+    return {"accuracy": accuracy}

examples/flows/evaluation/eval-classification-accuracy-maf/data.jsonl

@@ -0,0 +1,3 @@
+{"groundtruth": "App","prediction": "App"}
+{"groundtruth": "Channel","prediction": "Channel"}
+{"groundtruth": "Academic","prediction": "Academic"}

examples/flows/evaluation/eval-classification-accuracy-maf/eval_runner.py

@@ -0,0 +1,54 @@
+import asyncio
+from dataclasses import dataclass, field
+from typing import Any, Callable, Dict, List, Optional
+
+
+@dataclass
+class EvalResult:
+    per_row_outputs: List[Any]
+    metrics: Dict[str, Any]
+    errors: List[tuple] = field(default_factory=list)
+
+
+class EvalRunner:
+    def __init__(self, workflow_factory: Callable[[], Any], aggregate_fn: Callable[..., dict],
+                 concurrency: int = 5, input_mapping: Optional[Dict[str, str]] = None):
+        self._workflow_factory = workflow_factory
+        self._aggregate_fn = aggregate_fn
+        self._concurrency = concurrency
+        self._input_mapping = input_mapping
+
+    async def run(self, dataset: List[Any]) -> EvalResult:
+        semaphore = asyncio.Semaphore(self._concurrency)
+        per_row_outputs: List[Any] = [None] * len(dataset)
+        errors: List[tuple] = []
+
+        async def _run_row(index: int, row: Any) -> None:
+            async with semaphore:
+                wf = self._workflow_factory()
+                result = await wf.run(row)
+                per_row_outputs[index] = result.get_outputs()[0]
+
+        tasks = [_run_row(i, row) for i, row in enumerate(dataset)]
+        results = await asyncio.gather(*tasks, return_exceptions=True)
+        succeeded_outputs: List[Any] = []
+        for i, r in enumerate(results):
+            if isinstance(r, Exception):
+                errors.append((i, r))
+            else:
+                succeeded_outputs.append(per_row_outputs[i])
+
+        aggregation_inputs = self._transpose(succeeded_outputs)
+        if self._input_mapping:
+            aggregation_inputs = {self._input_mapping.get(k, k): v for k, v in aggregation_inputs.items()}
+        metrics = self._aggregate_fn(**aggregation_inputs)
+        return EvalResult(per_row_outputs=succeeded_outputs, metrics=metrics, errors=errors)
+
+    @staticmethod
+    def _transpose(outputs: List[Any]) -> Dict[str, Any]:
+        if not outputs:
+            return {"values": []}
+        if not isinstance(outputs[0], dict):
+            return {"values": outputs}
+        keys = outputs[0].keys()
+        return {k: [o[k] for o in outputs] for k in keys}

examples/flows/evaluation/eval-classification-accuracy-maf/requirements.txt

@@ -0,0 +1 @@
+agent-framework>=1.0.1

examples/flows/evaluation/eval-classification-accuracy-maf/run_eval.py

@@ -0,0 +1,45 @@
+import argparse
+import asyncio
+import json
+from pathlib import Path
+from aggregation import aggregate
+from eval_runner import EvalRunner
+from workflow import EvalInput, create_workflow
+
+DEFAULT_DATA = Path(__file__).parent / "data.jsonl"
+
+
+def load_dataset(path: Path) -> list[EvalInput]:
+    rows = []
+    with open(path, encoding="utf-8") as f:
+        for line in f:
+            line = line.strip()
+            if line:
+                obj = json.loads(line)
+                rows.append(EvalInput(groundtruth=obj["groundtruth"], prediction=obj["prediction"]))
+    return rows
+
+
+async def main(data_path: Path, concurrency: int):
+    dataset = load_dataset(data_path)
+    print(f"Loaded {len(dataset)} rows from {data_path}")
+
+    runner = EvalRunner(
+        workflow_factory=create_workflow,
+        aggregate_fn=aggregate,
+        concurrency=concurrency,
+        input_mapping={"values": "grades"},
+    )
+    result = await runner.run(dataset)
+
+    print("\n--- Metrics ---")
+    for key, value in result.metrics.items():
+        print(f"  {key}: {value}")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--data", type=Path, default=DEFAULT_DATA)
+    parser.add_argument("--concurrency", type=int, default=5)
+    args = parser.parse_args()
+    asyncio.run(main(args.data, args.concurrency))

examples/flows/evaluation/eval-classification-accuracy-maf/test_eval.py

@@ -0,0 +1,65 @@
+import asyncio
+import json
+from pathlib import Path
+from aggregation import aggregate
+from eval_runner import EvalRunner
+from workflow import EvalInput, create_workflow
+
+
+async def test_correct():
+    wf = create_workflow()
+    result = await wf.run(EvalInput(groundtruth="APP", prediction="APP"))
+    assert result.get_outputs()[0] == "Correct"
+    print("PASS: test_correct")
+
+
+async def test_incorrect():
+    wf = create_workflow()
+    result = await wf.run(EvalInput(groundtruth="APP", prediction="WEB"))
+    assert result.get_outputs()[0] == "Incorrect"
+    print("PASS: test_incorrect")
+
+
+async def test_batch():
+    dataset = [
+        EvalInput(groundtruth="APP", prediction="APP"),
+        EvalInput(groundtruth="Channel", prediction="Channel"),
+        EvalInput(groundtruth="Academic", prediction="Finance"),
+    ]
+    runner = EvalRunner(
+        workflow_factory=create_workflow,
+        aggregate_fn=aggregate,
+        concurrency=5,
+        input_mapping={"values": "grades"},
+    )
+    result = await runner.run(dataset)
+    assert result.metrics["accuracy"] == 0.67
+    print("PASS: test_batch")
+
+
+async def test_data_jsonl():
+    """Run eval on every row in data.jsonl"""
+    data_path = Path(__file__).parent / "data.jsonl"
+    rows = [json.loads(line) for line in data_path.read_text(encoding="utf-8").splitlines() if line.strip()]
+    wf = create_workflow()
+    for i, row in enumerate(rows):
+        result = await wf.run(EvalInput(
+            groundtruth=row["groundtruth"],
+            prediction=row["prediction"],
+        ))
+        grade = result.get_outputs()[0]
+        assert grade in ("Correct", "Incorrect"), f"Row {i}: unexpected grade '{grade}'"
+        print(f"  Row {i}: grade={grade}")
+    print(f"PASS: test_data_jsonl ({len(rows)} rows)")
+
+
+async def main():
+    await test_correct()
+    await test_incorrect()
+    await test_batch()
+    await test_data_jsonl()
+    print("\nAll tests passed!")
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/flows/evaluation/eval-classification-accuracy-maf/workflow.py

@@ -0,0 +1,21 @@
+from dataclasses import dataclass
+from typing_extensions import Never
+from agent_framework import Executor, WorkflowBuilder, WorkflowContext, handler
+
+
+@dataclass
+class EvalInput:
+    groundtruth: str
+    prediction: str
+
+
+class GradeExecutor(Executor):
+    @handler
+    async def grade(self, input: EvalInput, ctx: WorkflowContext[Never, str]) -> None:
+        result = "Correct" if input.groundtruth.lower() == input.prediction.lower() else "Incorrect"
+        await ctx.yield_output(result)
+
+
+def create_workflow():
+    _grade = GradeExecutor(id="grade")
+    return WorkflowBuilder(name="EvalClassificationAccuracyRow", start_executor=_grade).build()