#!/usr/bin/env python3 """ Trajectory evaluation script - RecruitmentAssistant-H_A2A project Evaluates 6 trajectory metrics: 1. Exact match 2. In-order match 3. Any-order match 4. Precision 5. Recall 6. Single-tool use A2A_mix hybrid architecture notes: - LangGraph (Job Analysis): dynamic matching for batched tool execution - CrewAI (Candidate Evaluation): standard CrewAI structure - AutoGen (Interview Communication): ignore create_agent and keep only invoke_agent """ import os import re import yaml from pathlib import Path from typing import List, Dict, Tuple, Set from collections import defaultdict import pandas as pd import math from itertools import permutations, product class TrajectoryParser: """Parse execution_path.md and extract the full execution trajectory.""" def __init__(self, md_file_path: str, extract_types: List[str] = None): """ Args: md_file_path: Path to execution_path.md extract_types: Node types to extract. Default: ['Tool'] Options: 'SPAN', 'Chain', 'Tool', 'AGENT', 'LLM', 'Task Created', 'Crew Created' """ self.md_file_path = md_file_path self.extract_types = extract_types or ["Tool"] self.trajectory = [] def parse(self) -> List[str]: """Parse the file and return the execution trajectory sequence.""" if not os.path.exists(self.md_file_path): return [] with open(self.md_file_path, "r", encoding="utf-8") as f: content = f.read() # Extract the "Execution Path Tree" section (inside the code block) tree_match = re.search( r"## Execution Path Tree.*?```\n(.*?)```", content, re.DOTALL ) if not tree_match: return [] tree_content = tree_match.group(1) trajectory = [] for line in tree_content.split("\n"): # Remove tree-structure characters and keep the node content clean_line = re.sub(r"^[│├└─\s]+", "", line).strip() if not clean_line: continue # Remove error marker clean_line = re.sub(r"^❌\s+", "", clean_line) # Remove retry markers: (retry N) and [RETRYN] clean_line = re.sub(r"\s*\(retry\s+\d+\)", "", clean_line) clean_line = re.sub(r"\s*\[RETRY\d+\]", "", clean_line) # Remove ERROR info (keep node type/name; drop the [ERROR:...] part) clean_line = re.sub(r"\s*\[ERROR:[^\]]*\]", "", clean_line) # Extract node info node_info = self._extract_node_info(clean_line) if node_info and node_info["type"] in self.extract_types: trajectory.append(node_info["action"]) self.trajectory = trajectory return trajectory def _extract_node_info(self, line: str) -> dict: """Extract node information from one line. Note: the input line is expected to have removed the marker, retry markers, and ERROR info. Returns: {'type': str, 'action': str} or None """ # SPAN node: [SPAN] span_name [stats] # Use a looser match to tolerate residual special characters span_match = re.match(r"\[SPAN\]\s+([^\[]+?)(?:\s+\[.*?\])*\s*$", line) if span_match: span_name = span_match.group(1).strip() return {"type": "SPAN", "action": f"SPAN: {span_name}"} # Chain node: [Chain] chain_name [stats] # For Crew_xxx.kickoff, use wildcard Crew***.kickoff chain_match = re.match(r"\[Chain\]\s+([^\[]+?)(?:\s+\[.*?\])*\s*$", line) if chain_match: chain_name = chain_match.group(1).strip() # Wildcard normalization: Crew_UUID.kickoff -> Crew***.kickoff chain_name = re.sub( r"Crew_[a-f0-9\-]+\.kickoff", "Crew***.kickoff", chain_name ) return {"type": "Chain", "action": f"Chain: {chain_name}"} # AGENT node: [AGENT] agent_name._execute_core [stats] # A2A_mix special case: ignore AutoGen create_agent and keep only invoke_agent agent_match = re.match(r"\[AGENT\]\s+([^\[]+?)(?:\s+\[.*?\])*\s*$", line) if agent_match: agent_name = agent_match.group(1).strip() # Ignore create_agent (AutoGen-specific, excluded from evaluation) if agent_name.startswith("create_agent"): return None # Remove ._execute_core suffix agent_name = re.sub(r"\._execute_core$", "", agent_name) return {"type": "AGENT", "action": f"AGENT: {agent_name}"} # Tool node: [Tool] tool_name._use [time] tool_match = re.match( r"\[Tool\]\s+([^\[\]]+?)(?:\s+\[[\d.]+(?:ms|s)\])?(?:\s*@@@)?\s*$", line ) if tool_match: tool_name = tool_match.group(1).strip() # Remove ._use suffix tool_name = re.sub(r"\._use$", "", tool_name) return {"type": "Tool", "action": f"Tool: {tool_name}"} # LLM node: [LLM] model_name (tokens) [time] llm_match = re.match(r"\[LLM\]\s+([^\(\[]+)", line) if llm_match: model_name = llm_match.group(1).strip() return {"type": "LLM", "action": f"LLM: {model_name}"} # Task Created node: [Task Created] [time] task_match = re.match(r"\[Task Created\]", line) if task_match: return {"type": "Task Created", "action": "Task Created"} # Crew Created node: [Crew Created] [time] crew_match = re.match(r"\[Crew Created\]", line) if crew_match: return {"type": "Crew Created", "action": "Crew Created"} return None class TrajectoryEvaluator: """Trajectory evaluator implementing 6 metrics. A2A_mix: supports dynamic reference-trajectory selection. """ def __init__(self, reference_trajectory: List[str]): """ Args: reference_trajectory: Reference trajectory (ground truth) """ self.reference = reference_trajectory @staticmethod def detect_job_analysis_tools_pattern(predicted: List[str]) -> List[int]: """Detect the tools grouping pattern in the LangGraph Job Analysis stage. Returns a list such as [1,1,1] / [2,1] / [3], where each number is the count of Tool nodes under one Chain: tools block. Returns an empty list if detection fails or the pattern is invalid. """ # Locate Chain: LangGraph start_idx = -1 for i, s in enumerate(predicted): if s == "Chain: LangGraph": start_idx = i break if start_idx == -1: return [] # Count Chain: tools blocks and the unified_web_search under them group_counts = [] i = start_idx while i < len(predicted): s = predicted[i] # On Chain: tools, count unified_web_search under it if s == "Chain: tools": count = 0 i += 1 # Count until the next Chain or AGENT while i < len(predicted): if predicted[i].startswith("Chain: ") or predicted[i].startswith( "AGENT: " ): break if predicted[i] == "Tool: unified_web_search": count += 1 i += 1 if count > 0: group_counts.append(count) # Chain: format_output indicates the end of the LangGraph stage elif s == "Chain: format_output": break else: i += 1 # Validate grouping pattern if not group_counts: return [] total = sum(group_counts) # Accept only 3 or 4 searches if total not in [3, 4]: return [] # Validate grouping is within the allowed set if total == 3: allowed = {(1, 1, 1), (1, 2), (2, 1), (3,)} else: # total == 4 allowed = { (1, 1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2), (3, 1), (1, 3), (4,), } if tuple(group_counts) not in allowed: return [] return group_counts @staticmethod def build_job_analysis_reference_variant( base_reference: List[str], pattern: List[int] ) -> List[str]: """Build a Job Analysis reference variant based on a tools-grouping pattern. Args: base_reference: Base reference trajectory (3x or 4x) pattern: Tools grouping pattern. For example, [1,2] means 1 tool in the first group and 2 tools in the second group. Returns: The adjusted full reference trajectory """ try: # Locate Job Analysis start/end start_idx = base_reference.index("Chain: LangGraph") # Find Chain: format_output (end of Job Analysis) end_idx = -1 for i in range(start_idx, len(base_reference)): if base_reference[i] == "Chain: format_output": end_idx = i break if end_idx == -1: return base_reference # Split into three parts: before / Job Analysis / after before = base_reference[: start_idx + 1] # includes Chain: LangGraph after = base_reference[end_idx:] # starts from format_output # Build a new Job Analysis part job_analysis_part = [ "AGENT: agent", "LLM: *", "Chain: _should_continue", ] # Add Chain: tools and Tool nodes according to the pattern for count in pattern: job_analysis_part.append("Chain: tools") for _ in range(count): job_analysis_part.append("Tool: unified_web_search") # Add the second agent pass job_analysis_part.extend( [ "AGENT: agent", "LLM: *", "Chain: _should_continue", ] ) # Compose the full trajectory return before + job_analysis_part + after except (ValueError, IndexError): # If parsing fails, return the original reference return base_reference @staticmethod def detect_autogen_interview_tools(predicted: List[str]) -> List[str]: """Detect the tool list in the AutoGen Interview Communication stage. Returns: A list of tool strings (in order of appearance). Returns an empty list if not found. Example: ['Tool: execute_tool comprehensive_interview_material_generator', 'Tool: execute_tool email_template_generator'] """ # Locate invoke_agent interview_coordinator start_idx = -1 for i, s in enumerate(predicted): if s == "AGENT: invoke_agent interview_coordinator": start_idx = i break if start_idx == -1: return [] # Extract tools under invoke_agent tools = [] i = start_idx + 1 # Stop at the next SPAN (end of Interview Communication stage) while i < len(predicted): s = predicted[i] if s.startswith("SPAN: "): break if s.startswith("Tool: execute_tool "): tools.append(s) i += 1 return tools @staticmethod def generate_autogen_llm_patterns(num_tools: int) -> List[List[int]]: """Generate all possible AutoGen LLM-insertion patterns between tools. Args: num_tools: Number of tools Returns: A list of patterns. Each pattern is a list indicating whether to insert an LLM between each adjacent pair of tools. Example for num_tools=2: [[0], [1]] - [0] = Tool1 → Tool2 (no insertion) - [1] = Tool1 → LLM → Tool2 (insert 1 LLM) """ if num_tools < 2: return [[]] # fewer than 2 tools: no gaps # n tools have (n-1) gaps num_gaps = num_tools - 1 # Enumerate all 0/1 combinations: 2^(n-1) patterns = [] for i in range(2**num_gaps): pattern = [] for j in range(num_gaps): # Extract the j-th bit (0 or 1) pattern.append((i >> j) & 1) patterns.append(pattern) return patterns @staticmethod def build_autogen_interview_reference_variant( base_reference: List[str], llm_pattern: List[int] ) -> List[str]: """Build an Interview Communication reference variant based on an LLM-insertion pattern. Args: base_reference: Base reference trajectory llm_pattern: LLM insertion pattern. For example, [0] means no LLM between tools, [1] means insert one LLM. Returns: The adjusted full reference trajectory """ try: # Locate invoke_agent interview_coordinator start_idx = base_reference.index( "AGENT: invoke_agent interview_coordinator" ) # Find the end of this stage (next SPAN or end-of-list) end_idx = len(base_reference) for i in range(start_idx + 1, len(base_reference)): if base_reference[i].startswith("SPAN: "): end_idx = i break # Extract tools from the base reference tools = [] for i in range(start_idx + 1, end_idx): if base_reference[i].startswith("Tool: execute_tool "): tools.append(base_reference[i]) if not tools: return base_reference # Split into three parts: before / Interview Communication / after before = base_reference[: start_idx + 1] # includes AGENT after = base_reference[end_idx:] # starts from the next SPAN # Build Interview Communication part based on llm_pattern interview_part = ["LLM: *"] # leading LLM for i, tool in enumerate(tools): interview_part.append(tool) # If not the last tool, decide whether to insert an LLM if i < len(tools) - 1 and i < len(llm_pattern): if llm_pattern[i] == 1: interview_part.append("LLM: *") interview_part.append("LLM: *") # trailing LLM # Compose full trajectory return before + interview_part + after except (ValueError, IndexError): # If parsing fails, return the original reference return base_reference def _match_action(self, predicted_action: str, reference_action: str) -> bool: """ Match two actions with wildcard support. Args: predicted_action: Observed action reference_action: Reference action (may contain wildcards) Returns: True if match, False otherwise """ # Exact match if predicted_action == reference_action: return True # Wildcard match: "LLM: *" matches any "LLM: " if reference_action == "LLM: *" and predicted_action.startswith("LLM: "): return True return False def exact_match(self, predicted: List[str]) -> int: """ Exact match: the predicted trajectory must be identical to the reference (with wildcard support). Returns: 1 if exact match, 0 otherwise """ if len(predicted) != len(self.reference): return 0 for i in range(len(predicted)): if not self._match_action(predicted[i], self.reference[i]): return 0 return 1 def in_order_match(self, predicted: List[str]) -> int: """ In-order match: the reference must be a subsequence of the predicted trajectory (with wildcard support). Extra actions are allowed, but required steps must appear in order. Returns: 1 if in-order match, 0 otherwise """ if not self.reference: return 1 # empty reference always matches ref_idx = 0 for pred_action in predicted: if ref_idx < len(self.reference) and self._match_action( pred_action, self.reference[ref_idx] ): ref_idx += 1 # Check whether all reference steps were found in order return 1 if ref_idx == len(self.reference) else 0 def any_order_match(self, predicted: List[str]) -> int: """ Any-order match: the predicted trajectory must contain all required actions (with wildcard support). Order is ignored; extra actions are allowed. Returns: 1 if any-order match, 0 otherwise """ if not self.reference: return 1 # Copy predicted actions for matching pred_remaining = predicted.copy() # For each reference action, try to find a match in the prediction for ref_action in self.reference: matched = False for i, pred_action in enumerate(pred_remaining): if self._match_action(pred_action, ref_action): pred_remaining.pop(i) # remove matched action matched = True break if not matched: return 0 # a reference action was not matched return 1 def precision(self, predicted: List[str]) -> float: """ Precision: fraction of predicted actions that are considered correct by the reference (with wildcard support). Precision = TP / (TP + FP) TP: Correct actions in prediction FP: Incorrect/extra actions in prediction Returns: precision value (0.0 - 1.0) """ if not predicted: return 1.0 # no prediction -> no false positives if not self.reference: return 0.0 # empty reference but non-empty prediction -> all are incorrect # Copy reference actions for matching ref_remaining = self.reference.copy() tp = 0 # True Positives for pred_action in predicted: # Try to find a match in the reference for i, ref_action in enumerate(ref_remaining): if self._match_action(pred_action, ref_action): tp += 1 ref_remaining.pop(i) # remove matched action break fp = len(predicted) - tp # False Positives return tp / (tp + fp) if (tp + fp) > 0 else 0.0 def recall(self, predicted: List[str]) -> float: """ Recall: fraction of reference actions covered by the predicted trajectory (with wildcard support). Recall = TP / (TP + FN) TP: Covered required actions FN: Missing required actions Returns: recall value (0.0 - 1.0) """ if not self.reference: return 1.0 # empty reference -> nothing to recall if not predicted: return 0.0 # no prediction -> recall is 0 # Copy predicted actions for matching pred_remaining = predicted.copy() tp = 0 # True Positives for ref_action in self.reference: # Try to find a match in the prediction for i, pred_action in enumerate(pred_remaining): if self._match_action(pred_action, ref_action): tp += 1 pred_remaining.pop(i) # remove matched action break fn = len(self.reference) - tp # False Negatives return tp / (tp + fn) if (tp + fn) > 0 else 0.0 def single_tool_use(self, predicted: List[str], tool_name: str) -> int: """ Single-tool use: check whether a specific tool appears in the trajectory (with wildcard support). Args: predicted: Predicted trajectory tool_name: Target tool name Returns: 1 if tool is used, 0 otherwise """ # Check whether any predicted action matches the target tool for pred_action in predicted: if self._match_action(pred_action, tool_name): return 1 return 0 def evaluate_all( self, predicted: List[str], target_tools: List[str] = None ) -> Dict[str, float]: """ Evaluate all metrics. Args: predicted: Predicted trajectory target_tools: Tools to check for single-tool use Returns: A dictionary of metric results """ results = { "exact_match": self.exact_match(predicted), "in_order_match": self.in_order_match(predicted), "any_order_match": self.any_order_match(predicted), "precision": self.precision(predicted), "recall": self.recall(predicted), } # Single-tool use: average usage rate across all target tools if target_tools: tool_usage_count = sum( self.single_tool_use(predicted, tool) for tool in target_tools ) results["single_tool_use"] = ( tool_usage_count / len(target_tools) if target_tools else 0.0 ) return results class DatasetEvaluator: """Dataset-level evaluator.""" def __init__(self, config_file: str): """ Args: config_file: Path to the YAML config file that defines reference trajectories """ self.config_file = config_file self.config = self._load_config() self.reference_trajectory = self.config.get("reference_trajectory", []) # Dynamic reference trajectory: choose based on unified_web_search count self.reference_trajectory_3x = self.config.get("reference_trajectory_3x", None) self.reference_trajectory_4x = self.config.get("reference_trajectory_4x", None) self.use_dynamic_reference = ( self.reference_trajectory_3x is not None and self.reference_trajectory_4x is not None ) self.target_tools = self.config.get("target_tools", []) self.models = self.config.get("models", []) self.project_name = self.config.get( "project_name", "RecruitmentAssistant-H_A2A" ) # Extract types: default is Tool only; can be configured to include others self.extract_types = self.config.get("extract_types", ["Tool"]) # A2A_mix: enable dynamic matching for LangGraph batched tool execution # Enabled by default when dynamic reference is available self.enable_tools_pattern_matching = self.use_dynamic_reference # A2A_mix: enable dynamic matching for AutoGen Interview LLM/Tool patterns # Enabled by default when dynamic reference is available self.enable_autogen_pattern_matching = self.use_dynamic_reference # Permutable tool groups (for dynamic tool-order optimization) self.permutable_tool_groups = self.config.get("permutable_tool_groups", {}) def _load_config(self) -> Dict: """Load the YAML config file.""" if not os.path.exists(self.config_file): print(f"⚠️ Config file does not exist: {self.config_file}") return {} with open(self.config_file, "r", encoding="utf-8") as f: return yaml.safe_load(f) def _count_unified_web_search_in_analyze_job(self, exec_path_file: str) -> int: """ Count unified_web_search occurrences under the analyze_job SPAN. Args: exec_path_file: Path to execution_path.md Returns: Number of unified_web_search calls """ if not os.path.exists(exec_path_file): return 0 with open(exec_path_file, "r", encoding="utf-8") as f: content = f.read() # Extract Execution Path Tree tree_match = re.search( r"## Execution Path Tree.*?```\n(.*?)```", content, re.DOTALL ) if not tree_match: return 0 tree_content = tree_match.group(1) lines = tree_content.split("\n") # Locate the analyze_job SPAN range in_analyze_job = False analyze_job_level = -1 count = 0 for line in lines: # Compute current line level (via leading tree characters) level = len(re.match(r"^([│├└─\s]*)", line).group(1)) # Normalize line content clean_line = re.sub(r"^[│├└─\s]+", "", line).strip() clean_line = re.sub(r"^❌\s+", "", clean_line) # Enter analyze_job SPAN if "[SPAN] analyze_job" in clean_line: in_analyze_job = True analyze_job_level = level continue # While inside analyze_job SPAN if in_analyze_job: # If we hit a SPAN at the same/higher level, we've left analyze_job if "[SPAN]" in clean_line and level <= analyze_job_level: break # Count unified_web_search if "[Tool] unified_web_search" in clean_line: count += 1 return count def _generate_permuted_trajectories( self, base_trajectory: List[str] ) -> List[List[str]]: """ Generate all possible permuted trajectories based on permutable_tool_groups. Args: base_trajectory: Base reference trajectory Returns: A list of all possible permuted trajectories (including the original one) """ if not self.permutable_tool_groups: # No permutable tool groups configured; return the original trajectory return [base_trajectory] # Collect all permutable tool groups and their positions in the trajectory tool_groups_positions = [] for group_name, tools in self.permutable_tool_groups.items(): # Find positions of this tool group in the trajectory positions = [] tool_indices = {} for i, action in enumerate(base_trajectory): for tool in tools: if action == tool: positions.append(i) tool_indices[i] = tool break # Only permute if all tools are found if len(positions) == len(tools): # Record positions and tools for this group tools_at_positions = [tool_indices[pos] for pos in positions] tool_groups_positions.append((positions, tools_at_positions)) if not tool_groups_positions: # No complete permutable tool group found return [base_trajectory] # Generate all possible permutation combinations all_trajectories = [] # Generate permutations for each tool group group_permutations = [] for positions, tools in tool_groups_positions: # Generate all permutations for this tool group perms = list(permutations(tools)) group_permutations.append([(positions, perm) for perm in perms]) # Cartesian product: combine permutations across tool groups all_group_combinations = list(product(*group_permutations)) # Create a new trajectory for each combination for combination in all_group_combinations: new_trajectory = base_trajectory.copy() # Apply all tool permutations in this combination for positions, perm in combination: for pos, tool in zip(positions, perm): new_trajectory[pos] = tool all_trajectories.append(new_trajectory) return all_trajectories def _find_best_reference_trajectory( self, predicted: List[str], candidate_references: List[List[str]] ) -> Tuple[List[str], Dict[str, float]]: """ Select the best reference trajectory from multiple candidates. Args: predicted: Predicted trajectory candidate_references: Candidate reference trajectories Returns: (best reference trajectory, corresponding match-score dict) """ best_reference = candidate_references[0] best_score = -1 best_metrics = {} for ref_trajectory in candidate_references: evaluator = TrajectoryEvaluator(ref_trajectory) # Compute three key match metrics exact = evaluator.exact_match(predicted) in_order = evaluator.in_order_match(predicted) any_order = evaluator.any_order_match(predicted) # Weighted score: exact_match has the highest weight # score = exact*3 + in_order*2 + any_order*1 score = exact * 3 + in_order * 2 + any_order * 1 if score > best_score: best_score = score best_reference = ref_trajectory best_metrics = { "exact_match": exact, "in_order_match": in_order, "any_order_match": any_order, } return best_reference, best_metrics def collect_execution_paths( self, model_name: str, base_dir: str ) -> List[Tuple[str, List[str]]]: """ Collect and parse all execution_path.md files for a given model. Args: model_name: Model name base_dir: RESULTS directory path Returns: A list of (session_id, trajectory) """ model_dir = Path(base_dir) / model_name / self.project_name / "test_results" if not model_dir.exists(): print(f"⚠️ Model directory does not exist: {model_dir}") return [] results = [] # Iterate over session subdirectories for session_dir in sorted(model_dir.iterdir()): if not session_dir.is_dir(): continue exec_path_file = session_dir / "execution_path.md" if not exec_path_file.exists(): continue # Parse trajectory using configured extract types parser = TrajectoryParser( str(exec_path_file), extract_types=self.extract_types ) trajectory = parser.parse() results.append((session_dir.name, trajectory)) return results def evaluate_model(self, model_name: str, base_dir: str) -> Dict[str, float]: """ Evaluate one model across all samples. Args: model_name: Model name base_dir: RESULTS directory path Returns: A dictionary of averaged metrics """ model_dir = Path(base_dir) / model_name / self.project_name / "test_results" if not model_dir.exists(): print(f"⚠️ No trajectory data found for model {model_name}") return {} # Accumulate metrics across samples all_metrics = defaultdict(list) trajectories = [] # Iterate over sessions for session_dir in sorted(model_dir.iterdir()): if not session_dir.is_dir(): continue exec_path_file = session_dir / "execution_path.md" if not exec_path_file.exists(): continue # Parse trajectory parser = TrajectoryParser( str(exec_path_file), extract_types=self.extract_types ) predicted = parser.parse() # Dynamically select the reference trajectory if self.use_dynamic_reference: # Count unified_web_search occurrences in analyze_job search_count = self._count_unified_web_search_in_analyze_job( str(exec_path_file) ) # Choose the base reference trajectory based on the count if search_count <= 3: base_reference = self.reference_trajectory_3x else: base_reference = self.reference_trajectory_4x # A2A_mix: dynamic matching (LangGraph + AutoGen) if ( self.enable_tools_pattern_matching or self.enable_autogen_pattern_matching ): # 1) Detect LangGraph tools grouping pattern langgraph_pattern = None if self.enable_tools_pattern_matching: langgraph_pattern = ( TrajectoryEvaluator.detect_job_analysis_tools_pattern( predicted ) ) # 2) Detect tools in the AutoGen Interview stage autogen_tools = None if self.enable_autogen_pattern_matching: autogen_tools = ( TrajectoryEvaluator.detect_autogen_interview_tools( predicted ) ) # 3) Generate all possible reference-variant combinations # All possible LangGraph patterns if langgraph_pattern: total = sum(langgraph_pattern) if total == 3: langgraph_patterns = [[1, 1, 1], [1, 2], [2, 1], [3]] else: # total == 4 langgraph_patterns = [ [1, 1, 1, 1], [2, 1, 1], [1, 2, 1], [1, 1, 2], [3, 1], [1, 3], [4], ] else: langgraph_patterns = [None] # unchanged # All possible AutoGen LLM insertion patterns if autogen_tools: num_tools = len(autogen_tools) autogen_llm_patterns = ( TrajectoryEvaluator.generate_autogen_llm_patterns(num_tools) ) else: autogen_llm_patterns = [None] # unchanged # 4) Enumerate all combinations and select the highest-scoring one best_reference = base_reference best_score = -1 for lg_pat in langgraph_patterns: for ag_llm_pat in autogen_llm_patterns: # Build a variant variant = base_reference # Apply LangGraph variant if lg_pat is not None: variant = TrajectoryEvaluator.build_job_analysis_reference_variant( variant, lg_pat ) # Apply AutoGen LLM-insertion variant if ag_llm_pat is not None: variant = TrajectoryEvaluator.build_autogen_interview_reference_variant( variant, ag_llm_pat ) # Compute score test_evaluator = TrajectoryEvaluator(variant) exact = test_evaluator.exact_match(predicted) in_order = test_evaluator.in_order_match(predicted) any_order = test_evaluator.any_order_match(predicted) score = max(exact, in_order, any_order) # Keep the best-scoring variant if score > best_score: best_score = score best_reference = variant reference = best_reference else: reference = base_reference else: # Use the default reference trajectory reference = self.reference_trajectory # Tool-order permutation optimization (after LangGraph+AutoGen pattern matching) if self.permutable_tool_groups: # Generate all possible tool-order permutations candidate_references = self._generate_permuted_trajectories(reference) # Select the best reference trajectory from candidates if len(candidate_references) > 1: reference, _ = self._find_best_reference_trajectory( predicted, candidate_references ) # else: keep reference unchanged # Create evaluator and compute metrics evaluator = TrajectoryEvaluator(reference) metrics = evaluator.evaluate_all(predicted, self.target_tools) for key, value in metrics.items(): all_metrics[key].append(value) trajectories.append((session_dir.name, predicted)) # Compute averages avg_metrics = {} for key, values in all_metrics.items(): avg_metrics[key] = sum(values) / len(values) if values else 0.0 num_samples = len(trajectories) if num_samples > 0: path_counter = defaultdict(int) for session_id, predicted in trajectories: path_key = tuple(predicted) path_counter[path_key] += 1 unique_paths = len(path_counter) unique_path_ratio = unique_paths / num_samples if num_samples > 0 else 0.0 probs = [count / num_samples for count in path_counter.values()] H = -sum(p * math.log(p) for p in probs if p > 0) if len(probs) > 1: path_entropy = H / math.log(len(probs)) else: path_entropy = 0.0 else: unique_path_ratio = 0.0 path_entropy = 0.0 avg_metrics["unique_path_ratio"] = unique_path_ratio avg_metrics["path_entropy"] = path_entropy # Add sample count avg_metrics["num_samples"] = num_samples return avg_metrics def evaluate_all_models(self, base_dir: str = None) -> pd.DataFrame: """ Evaluate all models and generate a summary table. Args: base_dir: RESULTS directory path (default: two levels above this script) Returns: A DataFrame containing evaluation results for all models """ if base_dir is None: # Default path: two levels above this script base_dir = Path(__file__).parent.parent.parent results = [] for model_name in self.models: print(f"\n📊 Evaluating model: {model_name}") metrics = self.evaluate_model(model_name, str(base_dir)) if metrics: metrics["model"] = model_name results.append(metrics) print(f" ✅ Done, samples: {metrics['num_samples']}") else: print(" ❌ Skipped (no data)") if not results: print("\n❌ No model data found") return pd.DataFrame() # Create DataFrame df = pd.DataFrame(results) # Reorder columns: put model name first cols = [ "model", "num_samples", "exact_match", "in_order_match", "any_order_match", "precision", "recall", "single_tool_use", "unique_path_ratio", "path_entropy", ] # Keep only existing columns cols = [col for col in cols if col in df.columns] df = df[cols] return df def main(): """Main entry point.""" import argparse parser = argparse.ArgumentParser( description="Evaluate trajectory metrics for the RecruitmentAssistant-H_A2A project" ) parser.add_argument( "--config", type=str, default="reference_trajectory.yaml", help="Path to the reference-trajectory config file (YAML)", ) parser.add_argument( "--base-dir", type=str, default=None, help="Path to the RESULTS directory (default: two levels above this script)", ) parser.add_argument( "--output", type=str, default="evaluation_results.csv", help="Output CSV file path", ) args = parser.parse_args() # If config path is relative, resolve it against the script directory config_path = args.config if not os.path.isabs(config_path): config_path = os.path.join(os.path.dirname(__file__), config_path) print("=" * 80) print("Trajectory Evaluation Tool - RecruitmentAssistant-H_A2A") print("=" * 80) print(f"\n📁 Config file: {config_path}") # Create evaluator evaluator = DatasetEvaluator(config_path) print(f"📋 Project: {evaluator.project_name}") # Reference trajectory info if evaluator.use_dynamic_reference: print("🎯 Reference trajectory: dynamic selection") print( f" - 3x version (unified_web_search<=3): {len(evaluator.reference_trajectory_3x)} steps" ) print( f" - 4x version (unified_web_search>=4): {len(evaluator.reference_trajectory_4x)} steps" ) else: print(f"🎯 Reference trajectory: {evaluator.reference_trajectory}") print(f"🔧 Target tools: {evaluator.target_tools}") print(f"🤖 Models: {evaluator.models}") # Dynamic optimization info if evaluator.use_dynamic_reference: print("\n🧠 Dynamic reference-trajectory optimization: enabled (3 layers)") print(" - Layer 1: LangGraph tools batching pattern") print(" * 3 searches: 4 grouping patterns") print(" * 4 searches: 7 grouping patterns") print(" - Layer 2: AutoGen LLM insertion pattern between tools") print(" * 2 tools: 2^1 = 2 patterns (0 or 1 LLM between tools)") print(" * 3 tools: 2^2 = 4 patterns") print(" - Layer 3: tool-call order permutation optimization") # Permutation info if evaluator.permutable_tool_groups: if not evaluator.use_dynamic_reference: print("\n🔀 Tool-order permutation optimization: enabled") total_permutations = 1 for group_name, tools in evaluator.permutable_tool_groups.items(): num_perms = math.factorial(len(tools)) total_permutations *= num_perms print(f" * {group_name}: {len(tools)} tools, {num_perms} permutations") print( f" - Total combinations (upper bound): 7 × 2 × {total_permutations} = {7 * 2 * total_permutations}" ) print( " - Strategy: select the best-matching reference combination per sample" ) else: if not evaluator.use_dynamic_reference: print("\n🔀 Tool-order permutation optimization: disabled") # Evaluate all models df = evaluator.evaluate_all_models(args.base_dir) if df.empty: print("\n❌ Evaluation failed: no data") return print("\n" + "=" * 80) print("📊 Evaluation results") print("=" * 80) # Display formatting pd.set_option("display.max_columns", None) pd.set_option("display.width", None) pd.set_option("display.float_format", lambda x: f"{x:.4f}") print("\n" + df.to_string(index=False)) # Save results output_dir = os.path.dirname(args.output) or "." os.makedirs(output_dir, exist_ok=True) csv_file = args.output df.to_csv(csv_file, index=False) print(f"\n✅ CSV saved: {csv_file}") print("\n" + "=" * 80) print("✅ Done") print("=" * 80) if __name__ == "__main__": main()