Introduce Evaluator Experiment (#3133)

* feat: add exchanged messages in LLMCallCompletedEvent

* feat: add GoalAlignment metric for Agent evaluation

* feat: add SemanticQuality metric for Agent evaluation

* feat: add Tool Metrics for Agent evaluation

* feat: add Reasoning Metrics for Agent evaluation, still in progress

* feat: add AgentEvaluator class

This class will evaluate Agent' results and report to user

* fix: do not evaluate Agent by default

This is a experimental feature we still need refine it further

* test: add Agent eval tests

* fix: render all feedback per iteration

* style: resolve linter issues

* style: fix mypy issues

* fix: allow messages be empty on LLMCallCompletedEvent

* feat: add Experiment evaluation framework with baseline comparison

* fix: reset evaluator for each experiement iteraction

* fix: fix track of new test cases

* chore: split Experimental evaluation classes

* refactor: remove unused method

* refactor: isolate Console print in a dedicated class

* fix: make crew required to run an experiment

* fix: use time-aware to define experiment result

* test: add tests for Evaluator Experiment

* style: fix linter issues

* fix: encode string before hashing

* style: resolve linter issues

* feat: add experimental folder for beta features (#3141)

* test: move tests to experimental folder

tests/experimental/evaluation/metrics/test_goal_metrics.py

@@ -1,8 +1,8 @@
 from unittest.mock import patch, MagicMock
-from tests.evaluation.metrics.base_evaluation_metrics_test import BaseEvaluationMetricsTest
+from tests.experimental.evaluation.metrics.base_evaluation_metrics_test import BaseEvaluationMetricsTest
 
-from crewai.evaluation.base_evaluator import EvaluationScore
-from crewai.evaluation.metrics.goal_metrics import GoalAlignmentEvaluator
+from crewai.experimental.evaluation.base_evaluator import EvaluationScore
+from crewai.experimental.evaluation.metrics.goal_metrics import GoalAlignmentEvaluator
 from crewai.utilities.llm_utils import LLM
 
 

tests/experimental/evaluation/metrics/test_reasoning_metrics.py

@@ -3,12 +3,12 @@ from unittest.mock import patch, MagicMock
 from typing import List, Dict, Any
 
 from crewai.tasks.task_output import TaskOutput
-from crewai.evaluation.metrics.reasoning_metrics import (
+from crewai.experimental.evaluation.metrics.reasoning_metrics import (
     ReasoningEfficiencyEvaluator,
 )
-from tests.evaluation.metrics.base_evaluation_metrics_test import BaseEvaluationMetricsTest
+from tests.experimental.evaluation.metrics.base_evaluation_metrics_test import BaseEvaluationMetricsTest
 from crewai.utilities.llm_utils import LLM
-from crewai.evaluation.base_evaluator import EvaluationScore
+from crewai.experimental.evaluation.base_evaluator import EvaluationScore
 
 class TestReasoningEfficiencyEvaluator(BaseEvaluationMetricsTest):
     @pytest.fixture

tests/experimental/evaluation/metrics/test_semantic_quality_metrics.py

@@ -1,8 +1,8 @@
 from unittest.mock import patch, MagicMock
 
-from crewai.evaluation.base_evaluator import EvaluationScore
-from crewai.evaluation.metrics.semantic_quality_metrics import SemanticQualityEvaluator
-from tests.evaluation.metrics.base_evaluation_metrics_test import BaseEvaluationMetricsTest
+from crewai.experimental.evaluation.base_evaluator import EvaluationScore
+from crewai.experimental.evaluation.metrics.semantic_quality_metrics import SemanticQualityEvaluator
+from tests.experimental.evaluation.metrics.base_evaluation_metrics_test import BaseEvaluationMetricsTest
 from crewai.utilities.llm_utils import LLM
 
 class TestSemanticQualityEvaluator(BaseEvaluationMetricsTest):

tests/experimental/evaluation/metrics/test_tools_metrics.py

@@ -1,12 +1,12 @@
 from unittest.mock import patch, MagicMock
 
-from crewai.evaluation.metrics.tools_metrics import (
+from crewai.experimental.evaluation.metrics.tools_metrics import (
     ToolSelectionEvaluator,
     ParameterExtractionEvaluator,
     ToolInvocationEvaluator
 )
 from crewai.utilities.llm_utils import LLM
-from tests.evaluation.metrics.base_evaluation_metrics_test import BaseEvaluationMetricsTest
+from tests.experimental.evaluation.metrics.base_evaluation_metrics_test import BaseEvaluationMetricsTest
 
 class TestToolSelectionEvaluator(BaseEvaluationMetricsTest):
     def test_no_tools_available(self, mock_task, mock_agent):

tests/experimental/evaluation/test_agent_evaluator.py

@@ -3,9 +3,9 @@ import pytest
 from crewai.agent import Agent
 from crewai.task import Task
 from crewai.crew import Crew
-from crewai.evaluation.agent_evaluator import AgentEvaluator
-from crewai.evaluation.base_evaluator import AgentEvaluationResult
-from crewai.evaluation import (
+from crewai.experimental.evaluation.agent_evaluator import AgentEvaluator
+from crewai.experimental.evaluation.base_evaluator import AgentEvaluationResult
+from crewai.experimental.evaluation import (
     GoalAlignmentEvaluator,
     SemanticQualityEvaluator,
     ToolSelectionEvaluator,
@@ -14,7 +14,7 @@ from crewai.evaluation import (
     ReasoningEfficiencyEvaluator
 )
 
-from crewai.evaluation import create_default_evaluator
+from crewai.experimental.evaluation import create_default_evaluator
 class TestAgentEvaluator:
     @pytest.fixture
     def mock_crew(self):

tests/experimental/evaluation/test_experiment_result.py

@@ -0,0 +1,111 @@
+import pytest
+from unittest.mock import MagicMock, patch
+
+from crewai.experimental.evaluation.experiment.result import ExperimentResult, ExperimentResults
+
+
+class TestExperimentResult:
+    @pytest.fixture
+    def mock_results(self):
+        return [
+            ExperimentResult(
+                identifier="test-1",
+                inputs={"query": "What is the capital of France?"},
+                score=10,
+                expected_score=7,
+                passed=True
+            ),
+            ExperimentResult(
+                identifier="test-2",
+                inputs={"query": "Who wrote Hamlet?"},
+                score={"relevance": 9, "factuality": 8},
+                expected_score={"relevance": 7, "factuality": 7},
+                passed=True,
+                agent_evaluations={"agent1": {"metrics": {"goal_alignment": {"score": 9}}}}
+            ),
+            ExperimentResult(
+                identifier="test-3",
+                inputs={"query": "Any query"},
+                score={"relevance": 9, "factuality": 8},
+                expected_score={"relevance": 7, "factuality": 7},
+                passed=False,
+                agent_evaluations={"agent1": {"metrics": {"goal_alignment": {"score": 9}}}}
+            ),
+            ExperimentResult(
+                identifier="test-4",
+                inputs={"query": "Another query"},
+                score={"relevance": 9, "factuality": 8},
+                expected_score={"relevance": 7, "factuality": 7},
+                passed=True,
+                agent_evaluations={"agent1": {"metrics": {"goal_alignment": {"score": 9}}}}
+            ),
+            ExperimentResult(
+                identifier="test-6",
+                inputs={"query": "Yet another query"},
+                score={"relevance": 9, "factuality": 8},
+                expected_score={"relevance": 7, "factuality": 7},
+                passed=True,
+                agent_evaluations={"agent1": {"metrics": {"goal_alignment": {"score": 9}}}}
+            )
+        ]
+
+    @patch('os.path.exists', return_value=True)
+    @patch('os.path.getsize', return_value=1)
+    @patch('json.load')
+    @patch('builtins.open', new_callable=MagicMock)
+    def test_experiment_results_compare_with_baseline(self, mock_open, mock_json_load, mock_path_getsize, mock_path_exists, mock_results):
+        baseline_data = {
+            "timestamp": "2023-01-01T00:00:00+00:00",
+            "results": [
+                {
+                    "identifier": "test-1",
+                    "inputs": {"query": "What is the capital of France?"},
+                    "score": 7,
+                    "expected_score": 7,
+                    "passed": False
+                },
+                {
+                    "identifier": "test-2",
+                    "inputs": {"query": "Who wrote Hamlet?"},
+                    "score": {"relevance": 8, "factuality": 7},
+                    "expected_score": {"relevance": 7, "factuality": 7},
+                    "passed": True
+                },
+                {
+                    "identifier": "test-3",
+                    "inputs": {"query": "Any query"},
+                    "score": {"relevance": 8, "factuality": 7},
+                    "expected_score": {"relevance": 7, "factuality": 7},
+                    "passed": True
+                },
+                {
+                    "identifier": "test-4",
+                    "inputs": {"query": "Another query"},
+                    "score": {"relevance": 8, "factuality": 7},
+                    "expected_score": {"relevance": 7, "factuality": 7},
+                    "passed": True
+                },
+                {
+                    "identifier": "test-5",
+                    "inputs": {"query": "Another query"},
+                    "score": {"relevance": 8, "factuality": 7},
+                    "expected_score": {"relevance": 7, "factuality": 7},
+                    "passed": True
+                }
+            ]
+        }
+
+        mock_json_load.return_value = baseline_data
+
+        results = ExperimentResults(results=mock_results)
+        results.display = MagicMock()
+
+        comparison = results.compare_with_baseline(baseline_filepath="baseline.json")
+
+        assert "baseline_timestamp" in comparison
+        assert comparison["baseline_timestamp"] == "2023-01-01T00:00:00+00:00"
+        assert comparison["improved"] == ["test-1"]
+        assert comparison["regressed"] == ["test-3"]
+        assert comparison["unchanged"] == ["test-2", "test-4"]
+        assert comparison["new_tests"] == ["test-6"]
+        assert comparison["missing_tests"] == ["test-5"]

tests/experimental/evaluation/test_experiment_runner.py

@@ -0,0 +1,197 @@
+import pytest
+from unittest.mock import MagicMock, patch
+
+from crewai.crew import Crew
+from crewai.experimental.evaluation.experiment.runner import ExperimentRunner
+from crewai.experimental.evaluation.experiment.result import ExperimentResults
+from crewai.experimental.evaluation.evaluation_display import AgentAggregatedEvaluationResult
+from crewai.experimental.evaluation.base_evaluator import MetricCategory, EvaluationScore
+
+
+class TestExperimentRunner:
+    @pytest.fixture
+    def mock_crew(self):
+        return MagicMock(llm=Crew)
+
+    @pytest.fixture
+    def mock_evaluator_results(self):
+        agent_evaluation = AgentAggregatedEvaluationResult(
+            agent_id="Test Agent",
+            agent_role="Test Agent Role",
+            metrics={
+                MetricCategory.GOAL_ALIGNMENT: EvaluationScore(
+                    score=9,
+                    feedback="Test feedback for goal alignment",
+                    raw_response="Test raw response for goal alignment"
+                ),
+                MetricCategory.REASONING_EFFICIENCY: EvaluationScore(
+                    score=None,
+                    feedback="Reasoning efficiency not applicable",
+                    raw_response="Reasoning efficiency not applicable"
+                ),
+                MetricCategory.PARAMETER_EXTRACTION: EvaluationScore(
+                    score=7,
+                    feedback="Test parameter extraction explanation",
+                    raw_response="Test raw output"
+                ),
+                MetricCategory.TOOL_SELECTION: EvaluationScore(
+                    score=8,
+                    feedback="Test tool selection explanation",
+                    raw_response="Test raw output"
+                )
+            }
+        )
+
+        return {"Test Agent": agent_evaluation}
+
+    @patch('crewai.experimental.evaluation.experiment.runner.create_default_evaluator')
+    def test_run_success(self, mock_create_evaluator, mock_crew, mock_evaluator_results):
+        dataset = [
+            {
+                "identifier": "test-case-1",
+                "inputs": {"query": "Test query 1"},
+                "expected_score": 8
+            },
+            {
+                "identifier": "test-case-2",
+                "inputs": {"query": "Test query 2"},
+                "expected_score": {"goal_alignment": 7}
+            },
+            {
+                "inputs": {"query": "Test query 3"},
+                "expected_score": {"tool_selection": 9}
+            }
+        ]
+
+        mock_evaluator = MagicMock()
+        mock_evaluator.get_agent_evaluation.return_value = mock_evaluator_results
+        mock_evaluator.reset_iterations_results = MagicMock()
+        mock_create_evaluator.return_value = mock_evaluator
+
+        runner = ExperimentRunner(dataset=dataset)
+
+        results = runner.run(crew=mock_crew)
+
+        assert isinstance(results, ExperimentResults)
+        result_1, result_2, result_3 = results.results
+        assert len(results.results) == 3
+
+        assert result_1.identifier == "test-case-1"
+        assert result_1.inputs == {"query": "Test query 1"}
+        assert result_1.expected_score == 8
+        assert result_1.passed is True
+
+        assert result_2.identifier == "test-case-2"
+        assert result_2.inputs == {"query": "Test query 2"}
+        assert isinstance(result_2.expected_score, dict)
+        assert "goal_alignment" in result_2.expected_score
+        assert result_2.passed is True
+
+        assert result_3.identifier == "c2ed49e63aa9a83af3ca382794134fd5"
+        assert result_3.inputs == {"query": "Test query 3"}
+        assert isinstance(result_3.expected_score, dict)
+        assert "tool_selection" in result_3.expected_score
+        assert result_3.passed is False
+
+        assert mock_crew.kickoff.call_count == 3
+        mock_crew.kickoff.assert_any_call(inputs={"query": "Test query 1"})
+        mock_crew.kickoff.assert_any_call(inputs={"query": "Test query 2"})
+        mock_crew.kickoff.assert_any_call(inputs={"query": "Test query 3"})
+
+        assert mock_evaluator.reset_iterations_results.call_count == 3
+        assert mock_evaluator.get_agent_evaluation.call_count == 3
+
+
+    @patch('crewai.experimental.evaluation.experiment.runner.create_default_evaluator')
+    def test_run_success_with_unknown_metric(self, mock_create_evaluator, mock_crew, mock_evaluator_results):
+        dataset = [
+            {
+                "identifier": "test-case-2",
+                "inputs": {"query": "Test query 2"},
+                "expected_score": {"goal_alignment": 7, "unknown_metric": 8}
+            }
+        ]
+
+        mock_evaluator = MagicMock()
+        mock_evaluator.get_agent_evaluation.return_value = mock_evaluator_results
+        mock_evaluator.reset_iterations_results = MagicMock()
+        mock_create_evaluator.return_value = mock_evaluator
+
+        runner = ExperimentRunner(dataset=dataset)
+
+        results = runner.run(crew=mock_crew)
+
+        result, = results.results
+
+        assert result.identifier == "test-case-2"
+        assert result.inputs == {"query": "Test query 2"}
+        assert isinstance(result.expected_score, dict)
+        assert "goal_alignment" in result.expected_score.keys()
+        assert "unknown_metric" in result.expected_score.keys()
+        assert result.passed is True
+
+    @patch('crewai.experimental.evaluation.experiment.runner.create_default_evaluator')
+    def test_run_success_with_single_metric_evaluator_and_expected_specific_metric(self, mock_create_evaluator, mock_crew, mock_evaluator_results):
+        dataset = [
+            {
+                "identifier": "test-case-2",
+                "inputs": {"query": "Test query 2"},
+                "expected_score": {"goal_alignment": 7}
+            }
+        ]
+
+        mock_evaluator = MagicMock()
+        mock_create_evaluator["Test Agent"].metrics = {
+            MetricCategory.GOAL_ALIGNMENT: EvaluationScore(
+                    score=9,
+                    feedback="Test feedback for goal alignment",
+                    raw_response="Test raw response for goal alignment"
+                )
+        }
+        mock_evaluator.get_agent_evaluation.return_value = mock_evaluator_results
+        mock_evaluator.reset_iterations_results = MagicMock()
+        mock_create_evaluator.return_value = mock_evaluator
+
+        runner = ExperimentRunner(dataset=dataset)
+
+        results = runner.run(crew=mock_crew)
+        result, = results.results
+
+        assert result.identifier == "test-case-2"
+        assert result.inputs == {"query": "Test query 2"}
+        assert isinstance(result.expected_score, dict)
+        assert "goal_alignment" in result.expected_score.keys()
+        assert result.passed is True
+
+    @patch('crewai.experimental.evaluation.experiment.runner.create_default_evaluator')
+    def test_run_success_when_expected_metric_is_not_available(self, mock_create_evaluator, mock_crew, mock_evaluator_results):
+        dataset = [
+            {
+                "identifier": "test-case-2",
+                "inputs": {"query": "Test query 2"},
+                "expected_score": {"unknown_metric": 7}
+            }
+        ]
+
+        mock_evaluator = MagicMock()
+        mock_create_evaluator["Test Agent"].metrics = {
+            MetricCategory.GOAL_ALIGNMENT: EvaluationScore(
+                score=5,
+                feedback="Test feedback for goal alignment",
+                raw_response="Test raw response for goal alignment"
+            )
+        }
+        mock_evaluator.get_agent_evaluation.return_value = mock_evaluator_results
+        mock_evaluator.reset_iterations_results = MagicMock()
+        mock_create_evaluator.return_value = mock_evaluator
+
+        runner = ExperimentRunner(dataset=dataset)
+
+        results = runner.run(crew=mock_crew)
+        result, = results.results
+
+        assert result.identifier == "test-case-2"
+        assert result.inputs == {"query": "Test query 2"}
+        assert isinstance(result.expected_score, dict)
+        assert "unknown_metric" in result.expected_score.keys()
+        assert result.passed is False