Function main

codeclash/analysis/viz/cdf_command_diversity.py:116–203

Main analysis function that processes all tournament data and generates the CDF plot. Process: 1. Scan all tournament directories for trajectory files 2. Extract commands from each trajectory and calculate diversity 3. Group diversity scores by model 4. Generate CDF visuali

()

def main():
    """
    Main analysis function that processes all tournament data and generates the CDF plot.

    Process:
    1. Scan all tournament directories for trajectory files
    2. Extract commands from each trajectory and calculate diversity
    3. Group diversity scores by model
    4. Generate CDF visualization comparing models
    """
    model_to_diversity = {}

    if not DATA_CACHE.exists():
        # Find all tournament directories by looking for metadata.json files
        tournaments = [x.parent for x in LOCAL_LOG_DIR.rglob("metadata.json")]
        for game_log_folder in tqdm(tournaments):
            # Load tournament metadata to get player-to-model mapping
            with open(game_log_folder / "metadata.json") as f:
                metadata = json.load(f)
            try:
                # Extract mapping from player name to model name
                p2m = {
                    x["name"]: x["config"]["model"]["model_name"].strip("@").split("/")[-1]
                    for x in metadata["config"]["players"]
                }
                # Initialize diversity list for each model we encounter
                for model in p2m.values():
                    if model not in model_to_diversity:
                        model_to_diversity[model] = []
            except KeyError:
                # Skip tournaments with malformed metadata
                continue

            # Process each player's trajectory files
            for name in p2m.keys():
                traj_files = (game_log_folder / "players" / name).rglob("*.traj.json")
                for traj_file in traj_files:
                    try:
                        with open(traj_file) as f:
                            traj = json.load(f)

                        # Extract commands and calculate diversity for this session
                        commands = extract_commands_from_trajectory(traj)
                        if commands:  # Only calculate entropy if there are commands
                            diversity = shannon_entropy(commands)
                            model_to_diversity[p2m[name]].append(diversity)
                    except (json.JSONDecodeError, KeyError):
                        # Skip malformed trajectory files
                        continue

        # Remove models with no valid data
        model_to_diversity = {k: v for k, v in model_to_diversity.items() if v}

        with open(DATA_CACHE, "w") as f:
            json.dump(model_to_diversity, f, indent=2)

    with open(DATA_CACHE) as f:
        model_to_diversity = json.load(f)