AINativeBench / data /processed /RQ1 /generate_radar_charts.py
王子睿
Update processed data
c4e7970 7 months ago
50.9 kB
#!/usr/bin/env python3
"""
Generate radar charts for evaluation results across different scenarios.
Creates one PDF radar chart per subfolder showing all models' performance.
"""

import pandas as pd
import numpy as np
from pathlib import Path
import math

try:
    import matplotlib.pyplot as plt
    import matplotlib as mpl
    from matplotlib.patches import Circle

    mpl.rcParams["font.family"] = "Times New Roman"
except ImportError:
    plt = None
    mpl = None
    Circle = None

# Metrics to plot
METRICS = [
    "exact_match",
    "any_order_match",
    "precision",
    "recall",
    "retry_rate",
    "pass_rate",
]

# Color palette for different models - professional academic colors (Tableau 10 style)
MODEL_COLORS = {
    "GPT-5": "#1f77b4",  # Blue
    "GPT-4o-mini": "#ff7f0e",  # Orange
    "DeepSeek-V3-1": "#2ca02c",  # Green
    "DeepSeek-R1": "#d62728",  # Red
    "Gemini-2.5-flash": "#9467bd",  # Purple
    "Gemini-2.5-flash-nothinking": "#8c564b",  # Brown
    "Qwen3-235b": "#e377c2",  # Pink
}

MODEL_CATEGORY = {
    "GPT-5": "closed",
    "GPT-4o-mini": "closed",
    "DeepSeek-V3-1": "open",
    "DeepSeek-R1": "open",
    "Gemini-2.5-flash": "closed",
    "Gemini-2.5-flash-nothinking": "closed",
    "Qwen3-235b": "open",
}


def df_to_markdown(df):
    """Convert a DataFrame to a simple markdown table without external deps."""
    headers = list(df.columns)
    lines = []
    lines.append("| " + " | ".join(headers) + " |")
    lines.append("| " + " | ".join(["---"] * len(headers)) + " |")
    for _, row in df.iterrows():
        values = []
        for val in row:
            if isinstance(val, float):
                values.append(f"{val:.2f}")
            else:
                values.append(str(val))
        lines.append("| " + " | ".join(values) + " |")
    return "\n".join(lines)


def metric_display_name(metric, base_project, gt_projects):
    if metric == "pass_rate" and gt_projects and base_project in gt_projects:
        return "average_score"
    return metric


def series_metric_headers(series_name, gt_projects):
    rate_label = (
        "average_score" if gt_projects and series_name in gt_projects else "pass_rate"
    )
    return [(rate_label if m == "pass_rate" else m) for m in METRICS]


def generate_model_pair_comparison(
    model_a, model_b, data_dict, comparison_name, gt_projects=None
):
    """Generate comparison table between two models across all scenarios."""
    lines = []
    lines.append(f"# {comparison_name}\n\n")

    # Group scenarios by base project name (remove -A2A, -A2A_mix, -MCP suffixes)
    project_series = {}
    for scenario_name in data_dict.keys():
        # Extract base project name
        base_name = scenario_name
        for suffix in ["-A2A_mix", "-A2A", "-MCP"]:
            if scenario_name.endswith(suffix):
                base_name = scenario_name[: -len(suffix)]
                break

        if base_name not in project_series:
            project_series[base_name] = []
        project_series[base_name].append(scenario_name)

    # Calculate averaged metrics for each project series
    series_data = {}
    for base_name, scenarios in project_series.items():
        series_data[base_name] = {model_a: {}, model_b: {}}

        for metric in METRICS:
            vals_a = []
            vals_b = []
            for scenario_name in scenarios:
                scenario_data = data_dict[scenario_name]
                if model_a in scenario_data and model_b in scenario_data:
                    vals_a.append(scenario_data[model_a].get(metric, 0))
                    vals_b.append(scenario_data[model_b].get(metric, 0))

            if vals_a and vals_b:
                series_data[base_name][model_a][metric] = np.mean(vals_a)
                series_data[base_name][model_b][metric] = np.mean(vals_b)

    # Calculate overall averages
    overall_avgs = {
        metric: {"a": [], "b": []} for metric in METRICS if metric != "pass_rate"
    }
    overall_rate_avgs = {
        "gt": {"a": [], "b": []},
        "non_gt": {"a": [], "b": []},
    }
    for base_name in series_data.keys():
        for metric in METRICS:
            if (
                metric not in series_data[base_name][model_a]
                or metric not in series_data[base_name][model_b]
            ):
                continue

            if metric == "pass_rate":
                bucket = "gt" if gt_projects and base_name in gt_projects else "non_gt"
                overall_rate_avgs[bucket]["a"].append(
                    series_data[base_name][model_a][metric]
                )
                overall_rate_avgs[bucket]["b"].append(
                    series_data[base_name][model_b][metric]
                )
                continue

            overall_avgs[metric]["a"].append(series_data[base_name][model_a][metric])
            overall_avgs[metric]["b"].append(series_data[base_name][model_b][metric])

    # Add overall summary section
    lines.append("## Overall Summary (Averaged Across All Project Series)\n\n")
    lines.append(f"| Metric | {model_a} | {model_b} | Diff (A-B) |\n")
    lines.append("| --- | --- | --- | --- |\n")
    for metric in METRICS:
        if metric == "pass_rate":
            if overall_rate_avgs["gt"]["a"]:
                avg_a = np.mean(overall_rate_avgs["gt"]["a"])
                avg_b = np.mean(overall_rate_avgs["gt"]["b"])
                diff = avg_a - avg_b
                lines.append(
                    f"| average_score | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n"
                )
            if overall_rate_avgs["non_gt"]["a"]:
                avg_a = np.mean(overall_rate_avgs["non_gt"]["a"])
                avg_b = np.mean(overall_rate_avgs["non_gt"]["b"])
                diff = avg_a - avg_b
                lines.append(
                    f"| pass_rate | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n"
                )
            continue

        if overall_avgs[metric]["a"]:
            avg_a = np.mean(overall_avgs[metric]["a"])
            avg_b = np.mean(overall_avgs[metric]["b"])
            diff = avg_a - avg_b
            lines.append(f"| {metric} | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n")
    lines.append("\n---\n\n")

    # Add per-series comparisons
    headers = ["Project Series", f"{model_a}", f"{model_b}", "Diff (A-B)"]

    for metric in METRICS:
        if metric != "pass_rate":
            lines.append(f"## {metric}\n\n")
            lines.append("| " + " | ".join(headers) + " |\n")
            lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")

            for base_name in sorted(series_data.keys()):
                if (
                    metric in series_data[base_name][model_a]
                    and metric in series_data[base_name][model_b]
                ):
                    val_a = series_data[base_name][model_a][metric]
                    val_b = series_data[base_name][model_b][metric]
                    diff = val_a - val_b
                    lines.append(
                        f"| {base_name} | {val_a:.2f} | {val_b:.2f} | {diff:+.2f} |\n"
                    )
            lines.append("\n")
            continue

        lines.append("## average_score\n\n")
        lines.append("| " + " | ".join(headers) + " |\n")
        lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")
        for base_name in sorted(series_data.keys()):
            if gt_projects and base_name not in gt_projects:
                continue
            if (
                metric in series_data[base_name][model_a]
                and metric in series_data[base_name][model_b]
            ):
                val_a = series_data[base_name][model_a][metric]
                val_b = series_data[base_name][model_b][metric]
                diff = val_a - val_b
                lines.append(
                    f"| {base_name} | {val_a:.2f} | {val_b:.2f} | {diff:+.2f} |\n"
                )
        lines.append("\n")

        lines.append("## pass_rate\n\n")
        lines.append("| " + " | ".join(headers) + " |\n")
        lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")
        for base_name in sorted(series_data.keys()):
            if gt_projects and base_name in gt_projects:
                continue
            if (
                metric in series_data[base_name][model_a]
                and metric in series_data[base_name][model_b]
            ):
                val_a = series_data[base_name][model_a][metric]
                val_b = series_data[base_name][model_b][metric]
                diff = val_a - val_b
                lines.append(
                    f"| {base_name} | {val_a:.2f} | {val_b:.2f} | {diff:+.2f} |\n"
                )
        lines.append("\n")

    return "".join(lines)


def generate_mcp_comparison(base_project, scenario_data_dict, gt_projects=None):
    """Generate comparison between MCP and hardcoded versions for a project."""
    lines = []
    lines.append(f"# {base_project}: MCP vs Hardcoded\n\n")

    mcp_scenario = f"{base_project}-MCP"
    hardcoded_scenario = base_project

    if (
        mcp_scenario not in scenario_data_dict
        or hardcoded_scenario not in scenario_data_dict
    ):
        lines.append(f"_Data not available for comparison_\n\n")
        return "".join(lines)

    mcp_data = scenario_data_dict[mcp_scenario]
    hardcoded_data = scenario_data_dict[hardcoded_scenario]

    all_models = set(mcp_data.keys()) | set(hardcoded_data.keys())

    # Calculate overall averages first
    overall_avgs = {metric: {"mcp": [], "hard": []} for metric in METRICS}
    for model in all_models:
        for metric in METRICS:
            mcp_val = mcp_data.get(model, {}).get(metric, 0)
            hard_val = hardcoded_data.get(model, {}).get(metric, 0)
            overall_avgs[metric]["mcp"].append(mcp_val)
            overall_avgs[metric]["hard"].append(hard_val)

    # Add overall summary section
    lines.append("## Overall Summary (Averaged Across All Models)\n\n")
    lines.append("| Metric | MCP | Hardcoded | Diff (MCP-Hardcoded) |\n")
    lines.append("| --- | --- | --- | --- |\n")
    for metric in METRICS:
        if overall_avgs[metric]["mcp"]:
            avg_mcp = np.mean(overall_avgs[metric]["mcp"])
            avg_hard = np.mean(overall_avgs[metric]["hard"])
            diff = avg_mcp - avg_hard
            display_metric = metric_display_name(metric, base_project, gt_projects)
            lines.append(
                f"| {display_metric} | {avg_mcp:.2f} | {avg_hard:.2f} | {diff:+.2f} |\n"
            )
    lines.append("\n---\n\n")

    for metric in METRICS:
        display_metric = metric_display_name(metric, base_project, gt_projects)
        lines.append(f"## {display_metric}\n\n")
        headers = ["Model", "MCP", "Hardcoded", "Diff (MCP-Hardcoded)"]
        lines.append("| " + " | ".join(headers) + " |\n")
        lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")

        for model in sorted(all_models):
            mcp_val = mcp_data.get(model, {}).get(metric, 0)
            hard_val = hardcoded_data.get(model, {}).get(metric, 0)
            diff = mcp_val - hard_val
            lines.append(
                f"| {model} | {mcp_val:.2f} | {hard_val:.2f} | {diff:+.2f} |\n"
            )
        lines.append("\n")

    return "".join(lines)


def generate_mcp_overall_comparison(projects, scenario_data_dict, gt_projects=None):
    """Generate overall comparison across all MCP vs hardcoded projects."""
    lines = []
    lines.append("# Overall MCP vs Hardcoded Comparison\n\n")
    lines.append(
        "Averaged across all projects: MarkdownValidator, GameBuilder, EmailResponder\n\n"
    )

    # Collect data from all projects
    overall_data = {}
    framework_data = {
        metric: {"mcp": [], "hard": []} for metric in METRICS if metric != "pass_rate"
    }
    framework_rate_data = {
        "gt": {"mcp": [], "hard": []},
        "non_gt": {"mcp": [], "hard": []},
    }
    overall_rate_data = {}

    for project in projects:
        mcp_scenario = f"{project}-MCP"
        hardcoded_scenario = project

        if (
            mcp_scenario not in scenario_data_dict
            or hardcoded_scenario not in scenario_data_dict
        ):
            continue

        mcp_data = scenario_data_dict[mcp_scenario]
        hardcoded_data = scenario_data_dict[hardcoded_scenario]

        all_models = set(mcp_data.keys()) | set(hardcoded_data.keys())
        project_bucket = "gt" if gt_projects and project in gt_projects else "non_gt"

        for model in all_models:
            if model not in overall_data:
                overall_data[model] = {
                    metric: {"mcp": [], "hard": []}
                    for metric in METRICS
                    if metric != "pass_rate"
                }
            if model not in overall_rate_data:
                overall_rate_data[model] = {
                    "gt": {"mcp": [], "hard": []},
                    "non_gt": {"mcp": [], "hard": []},
                }

            for metric in METRICS:
                mcp_val = mcp_data.get(model, {}).get(metric, 0)
                hard_val = hardcoded_data.get(model, {}).get(metric, 0)
                if metric == "pass_rate":
                    overall_rate_data[model][project_bucket]["mcp"].append(mcp_val)
                    overall_rate_data[model][project_bucket]["hard"].append(hard_val)
                    framework_rate_data[project_bucket]["mcp"].append(mcp_val)
                    framework_rate_data[project_bucket]["hard"].append(hard_val)
                else:
                    overall_data[model][metric]["mcp"].append(mcp_val)
                    overall_data[model][metric]["hard"].append(hard_val)
                    framework_data[metric]["mcp"].append(mcp_val)
                    framework_data[metric]["hard"].append(hard_val)

    # Add framework-level comparison (all models averaged)
    lines.append("## Framework-Level Comparison (All Models Averaged)\n\n")
    lines.append("| Metric | MCP | Hardcoded | Diff (MCP-Hardcoded) |\n")
    lines.append("| --- | --- | --- | --- |\n")
    for metric in METRICS:
        if metric == "pass_rate":
            if framework_rate_data["gt"]["mcp"]:
                avg_mcp = np.mean(framework_rate_data["gt"]["mcp"])
                avg_hard = np.mean(framework_rate_data["gt"]["hard"])
                diff = avg_mcp - avg_hard
                lines.append(
                    f"| average_score | {avg_mcp:.2f} | {avg_hard:.2f} | {diff:+.2f} |\n"
                )
            if framework_rate_data["non_gt"]["mcp"]:
                avg_mcp = np.mean(framework_rate_data["non_gt"]["mcp"])
                avg_hard = np.mean(framework_rate_data["non_gt"]["hard"])
                diff = avg_mcp - avg_hard
                lines.append(
                    f"| pass_rate | {avg_mcp:.2f} | {avg_hard:.2f} | {diff:+.2f} |\n"
                )
            continue

        if framework_data[metric]["mcp"]:
            avg_mcp = np.mean(framework_data[metric]["mcp"])
            avg_hard = np.mean(framework_data[metric]["hard"])
            diff = avg_mcp - avg_hard
            lines.append(
                f"| {metric} | {avg_mcp:.2f} | {avg_hard:.2f} | {diff:+.2f} |\n"
            )
    lines.append("\n---\n\n")

    # Generate per-model summary tables
    for metric in METRICS:
        if metric != "pass_rate":
            lines.append(f"## {metric}\n\n")
            headers = [
                "Model",
                "MCP (Avg)",
                "Hardcoded (Avg)",
                "Diff (MCP-Hardcoded)",
            ]
            lines.append("| " + " | ".join(headers) + " |\n")
            lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")

            for model in sorted(overall_data.keys()):
                if overall_data[model][metric]["mcp"]:
                    avg_mcp = np.mean(overall_data[model][metric]["mcp"])
                    avg_hard = np.mean(overall_data[model][metric]["hard"])
                    diff = avg_mcp - avg_hard
                    lines.append(
                        f"| {model} | {avg_mcp:.2f} | {avg_hard:.2f} | {diff:+.2f} |\n"
                    )
            lines.append("\n")
            continue

        lines.append("## average_score\n\n")
        headers = ["Model", "MCP (Avg)", "Hardcoded (Avg)", "Diff (MCP-Hardcoded)"]
        lines.append("| " + " | ".join(headers) + " |\n")
        lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")
        for model in sorted(overall_rate_data.keys()):
            if overall_rate_data[model]["gt"]["mcp"]:
                avg_mcp = np.mean(overall_rate_data[model]["gt"]["mcp"])
                avg_hard = np.mean(overall_rate_data[model]["gt"]["hard"])
                diff = avg_mcp - avg_hard
                lines.append(
                    f"| {model} | {avg_mcp:.2f} | {avg_hard:.2f} | {diff:+.2f} |\n"
                )
        lines.append("\n")

        lines.append("## pass_rate\n\n")
        lines.append("| " + " | ".join(headers) + " |\n")
        lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")
        for model in sorted(overall_rate_data.keys()):
            if overall_rate_data[model]["non_gt"]["mcp"]:
                avg_mcp = np.mean(overall_rate_data[model]["non_gt"]["mcp"])
                avg_hard = np.mean(overall_rate_data[model]["non_gt"]["hard"])
                diff = avg_mcp - avg_hard
                lines.append(
                    f"| {model} | {avg_mcp:.2f} | {avg_hard:.2f} | {diff:+.2f} |\n"
                )
        lines.append("\n")

    lines.append("---\n\n")
    return "".join(lines)


def generate_version_comparison(
    base_project,
    version_a_suffix,
    version_b_suffix,
    scenario_data_dict,
    version_a_name,
    version_b_name,
    gt_projects=None,
):
    """Generate comparison between two versions of a project."""
    lines = []
    lines.append(f"# {base_project}: {version_a_name} vs {version_b_name}\n\n")

    scenario_a = f"{base_project}{version_a_suffix}"
    scenario_b = f"{base_project}{version_b_suffix}"

    if scenario_a not in scenario_data_dict or scenario_b not in scenario_data_dict:
        lines.append(f"_Data not available for comparison_\n\n")
        return "".join(lines)

    data_a = scenario_data_dict[scenario_a]
    data_b = scenario_data_dict[scenario_b]

    all_models = set(data_a.keys()) | set(data_b.keys())

    # Calculate overall averages first
    overall_avgs = {metric: {"a": [], "b": []} for metric in METRICS}
    for model in all_models:
        for metric in METRICS:
            val_a = data_a.get(model, {}).get(metric, 0)
            val_b = data_b.get(model, {}).get(metric, 0)
            overall_avgs[metric]["a"].append(val_a)
            overall_avgs[metric]["b"].append(val_b)

    # Add overall summary section
    lines.append("## Overall Summary (Averaged Across All Models)\n\n")
    lines.append(
        f"| Metric | {version_a_name} | {version_b_name} | Diff ({version_a_name}-{version_b_name}) |\n"
    )
    lines.append("| --- | --- | --- | --- |\n")
    for metric in METRICS:
        if overall_avgs[metric]["a"]:
            avg_a = np.mean(overall_avgs[metric]["a"])
            avg_b = np.mean(overall_avgs[metric]["b"])
            diff = avg_a - avg_b
            display_metric = metric_display_name(metric, base_project, gt_projects)
            lines.append(
                f"| {display_metric} | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n"
            )
    lines.append("\n---\n\n")

    for metric in METRICS:
        display_metric = metric_display_name(metric, base_project, gt_projects)
        lines.append(f"## {display_metric}\n\n")
        headers = [
            "Model",
            version_a_name,
            version_b_name,
            f"Diff ({version_a_name}-{version_b_name})",
        ]
        lines.append("| " + " | ".join(headers) + " |\n")
        lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")

        for model in sorted(all_models):
            val_a = data_a.get(model, {}).get(metric, 0)
            val_b = data_b.get(model, {}).get(metric, 0)
            diff = val_a - val_b
            lines.append(f"| {model} | {val_a:.2f} | {val_b:.2f} | {diff:+.2f} |\n")
        lines.append("\n")

    return "".join(lines)


def generate_version_overall_comparison(
    projects,
    version_a_suffix,
    version_b_suffix,
    scenario_data_dict,
    version_a_name,
    version_b_name,
    comparison_title,
    gt_projects=None,
):
    """Generate overall comparison across all projects for two versions."""
    lines = []
    lines.append(f"# Overall {comparison_title}\n\n")
    lines.append(f"Averaged across all projects: {', '.join(projects)}\n\n")

    # Collect data from all projects
    overall_data = {}
    framework_data = {
        metric: {"a": [], "b": []} for metric in METRICS if metric != "pass_rate"
    }
    framework_rate_data = {
        "gt": {"a": [], "b": []},
        "non_gt": {"a": [], "b": []},
    }
    overall_rate_data = {}

    for project in projects:
        scenario_a = f"{project}{version_a_suffix}"
        scenario_b = f"{project}{version_b_suffix}"

        if scenario_a not in scenario_data_dict or scenario_b not in scenario_data_dict:
            continue

        data_a = scenario_data_dict[scenario_a]
        data_b = scenario_data_dict[scenario_b]

        all_models = set(data_a.keys()) | set(data_b.keys())
        project_bucket = "gt" if gt_projects and project in gt_projects else "non_gt"

        for model in all_models:
            if model not in overall_data:
                overall_data[model] = {
                    metric: {"a": [], "b": []}
                    for metric in METRICS
                    if metric != "pass_rate"
                }
            if model not in overall_rate_data:
                overall_rate_data[model] = {
                    "gt": {"a": [], "b": []},
                    "non_gt": {"a": [], "b": []},
                }

            for metric in METRICS:
                val_a = data_a.get(model, {}).get(metric, 0)
                val_b = data_b.get(model, {}).get(metric, 0)
                if metric == "pass_rate":
                    overall_rate_data[model][project_bucket]["a"].append(val_a)
                    overall_rate_data[model][project_bucket]["b"].append(val_b)
                    framework_rate_data[project_bucket]["a"].append(val_a)
                    framework_rate_data[project_bucket]["b"].append(val_b)
                else:
                    overall_data[model][metric]["a"].append(val_a)
                    overall_data[model][metric]["b"].append(val_b)
                    framework_data[metric]["a"].append(val_a)
                    framework_data[metric]["b"].append(val_b)

    # Add framework-level comparison (all models averaged)
    lines.append("## Framework-Level Comparison (All Models Averaged)\n\n")
    lines.append(
        f"| Metric | {version_a_name} | {version_b_name} | Diff ({version_a_name}-{version_b_name}) |\n"
    )
    lines.append("| --- | --- | --- | --- |\n")
    for metric in METRICS:
        if metric == "pass_rate":
            if framework_rate_data["gt"]["a"]:
                avg_a = np.mean(framework_rate_data["gt"]["a"])
                avg_b = np.mean(framework_rate_data["gt"]["b"])
                diff = avg_a - avg_b
                lines.append(
                    f"| average_score | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n"
                )
            if framework_rate_data["non_gt"]["a"]:
                avg_a = np.mean(framework_rate_data["non_gt"]["a"])
                avg_b = np.mean(framework_rate_data["non_gt"]["b"])
                diff = avg_a - avg_b
                lines.append(
                    f"| pass_rate | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n"
                )
            continue

        if framework_data[metric]["a"]:
            avg_a = np.mean(framework_data[metric]["a"])
            avg_b = np.mean(framework_data[metric]["b"])
            diff = avg_a - avg_b
            lines.append(f"| {metric} | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n")
    lines.append("\n---\n\n")

    # Generate per-model summary tables
    for metric in METRICS:
        headers = [
            "Model",
            f"{version_a_name} (Avg)",
            f"{version_b_name} (Avg)",
            f"Diff ({version_a_name}-{version_b_name})",
        ]

        if metric != "pass_rate":
            lines.append(f"## {metric}\n\n")
            lines.append("| " + " | ".join(headers) + " |\n")
            lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")

            for model in sorted(overall_data.keys()):
                if overall_data[model][metric]["a"]:
                    avg_a = np.mean(overall_data[model][metric]["a"])
                    avg_b = np.mean(overall_data[model][metric]["b"])
                    diff = avg_a - avg_b
                    lines.append(
                        f"| {model} | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n"
                    )
            lines.append("\n")
            continue

        lines.append("## average_score\n\n")
        lines.append("| " + " | ".join(headers) + " |\n")
        lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")
        for model in sorted(overall_rate_data.keys()):
            if overall_rate_data[model]["gt"]["a"]:
                avg_a = np.mean(overall_rate_data[model]["gt"]["a"])
                avg_b = np.mean(overall_rate_data[model]["gt"]["b"])
                diff = avg_a - avg_b
                lines.append(f"| {model} | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n")
        lines.append("\n")

        lines.append("## pass_rate\n\n")
        lines.append("| " + " | ".join(headers) + " |\n")
        lines.append("| " + " | ".join(["---"] * len(headers)) + " |\n")
        for model in sorted(overall_rate_data.keys()):
            if overall_rate_data[model]["non_gt"]["a"]:
                avg_a = np.mean(overall_rate_data[model]["non_gt"]["a"])
                avg_b = np.mean(overall_rate_data[model]["non_gt"]["b"])
                diff = avg_a - avg_b
                lines.append(f"| {model} | {avg_a:.2f} | {avg_b:.2f} | {diff:+.2f} |\n")
        lines.append("\n")

    lines.append("---\n\n")
    return "".join(lines)


def create_radar_chart(df, scenario_name, output_path, pass_rate_label="pass_rate"):
    """
    Create a radar chart for a single scenario with all models.
    Args:
        df: DataFrame containing evaluation results
        scenario_name: Name of the scenario (subfolder name)
        output_path: Path to save the PDF file
    """
    if plt is None:
        print(
            f" Warning: matplotlib not installed, skipping radar chart: {output_path}"
        )
        return
    # Number of metrics
    num_vars = len(METRICS)

    # Compute angle for each axis
    angles = np.linspace(0, 2 * np.pi, num_vars, endpoint=False).tolist()
    angles += angles[:1]  # Complete the circle

    # Create figure with polar projection - increase size to avoid overlap
    fig, ax = plt.subplots(figsize=(12, 12), subplot_kw=dict(projection="polar"))

    # Set background color to light blue/grey similar to reference
    ax.set_facecolor("#cedbea")  # slightly lighter blue background
    fig.patch.set_facecolor("white")

    # Plot data for each model
    for idx, row in df.iterrows():
        model_name = row["model"]
        values = [row[metric] for metric in METRICS]
        values += values[:1]  # Complete the circle

        color = MODEL_COLORS.get(model_name, f"C{idx}")

        # Plot the line and fill
        ax.plot(
            angles,
            values,
            "o-",
            linewidth=2,
            label=model_name,
            color=color,
            markersize=6,
        )
        ax.fill(angles, values, alpha=0.15, color=color)

    # Set the labels for each axis so they sit just outside the outer circle
    metric_labels = [
        (pass_rate_label if metric == "pass_rate" else metric) for metric in METRICS
    ]
    ax.set_xticks(angles[:-1])
    # Remove default tick labels; we'll draw them manually at a fixed radius
    ax.set_xticklabels([])

    label_radius = 1.3  # Place labels on the outer grid circle (radius=1.25)
    for angle, label in zip(angles[:-1], metric_labels):
        # Choose horizontal alignment based on angle for better spacing
        if 0 <= angle < np.pi / 2 or 3 * np.pi / 2 <= angle <= 2 * np.pi:
            ha = "left"
        elif np.pi / 2 < angle < 3 * np.pi / 2:
            ha = "right"
        else:
            ha = "center"
        ax.text(
            angle,
            label_radius,
            label,
            ha=ha,
            va="center",
            fontsize=30,
            fontweight="bold",
        )

    # Set radial limits to 1.25; keep labels only up to 1.0
    ax.set_ylim(0, 1.25)
    ax.set_yticks([0, 0.25, 0.5, 0.75, 1.0, 1.25])
    ax.set_yticklabels(
        ["0", "0.25", "0.5", "0.75", "1.0", ""],
        size=30,
        color="black",
        weight="bold",
    )

    # Remove the outer spine (circular border)
    ax.spines["polar"].set_visible(False)

    # Add gridlines
    ax.xaxis.grid(True, linestyle="--", linewidth=3.0, alpha=1.0, color="white")
    ax.yaxis.grid(True, linestyle="-", linewidth=3.0, color="white", alpha=1.0)

    # Title removed per user request
    # Adjust layout
    plt.tight_layout()

    # Save the figure
    plt.savefig(output_path, format="pdf", dpi=300, bbox_inches="tight")
    plt.close()

    print(f" Created: {output_path}")


def create_legend_pdf(output_path):
    if plt is None:
        print(f" Warning: matplotlib not installed, skipping legend PDF: {output_path}")
        return
    fig, ax = plt.subplots(figsize=(4, 3))
    ax.axis("off")

    handles = []
    labels = []
    for model_name, color in MODEL_COLORS.items():
        (handle,) = ax.plot(
            [],
            [],
            "o-",
            linewidth=2,
            markersize=8,
            color=color,
        )
        handles.append(handle)
        labels.append(model_name)

    ax.legend(
        handles,
        labels,
        loc="center",
        frameon=True,
        fancybox=True,
        shadow=False,
        prop={"size": 12, "weight": "bold"},
    )

    plt.tight_layout()
    plt.savefig(output_path, format="pdf", dpi=300, bbox_inches="tight")
    plt.close(fig)

    print(f" Legend PDF created: {output_path}")


def create_legend_pdf_horizontal(output_path):
    if plt is None:
        print(
            f" Warning: matplotlib not installed, skipping horizontal legend PDF: {output_path}"
        )
        return
    # Wide and very low-height figure to make the legend strip as "flat" as possible
    fig, ax = plt.subplots(figsize=(12, 1.0))
    ax.axis("off")

    handles = []
    labels = []
    for model_name, color in MODEL_COLORS.items():
        (handle,) = ax.plot(
            [],
            [],
            "o-",
            linewidth=2,
            markersize=8,
            color=color,
        )
        handles.append(handle)
        labels.append(model_name)

    ax.legend(
        handles,
        labels,
        loc="center",
        ncol=len(MODEL_COLORS),
        frameon=False,
        fancybox=False,
        shadow=False,
        borderaxespad=0.1,
        borderpad=0.3,
        handletextpad=0.4,
        labelspacing=0.2,
        prop={"size": 12, "weight": "bold"},
    )

    plt.tight_layout(pad=0.0)
    plt.savefig(output_path, format="pdf", dpi=300, bbox_inches="tight", pad_inches=0.0)
    plt.close(fig)

    print(f" Horizontal legend PDF created: {output_path}")


def main():
    """Main function to process all subfolders and generate radar charts."""
    base_dir = Path("/Users/wzr/TOSEM-2025/RESULTS/RQ1")
    output_dir = base_dir / "RadarCharts"
    output_dir.mkdir(exist_ok=True)

    # Find all evaluation_results.csv files
    csv_files = list(base_dir.glob("*/evaluation_results.csv"))

    print(f"Found {len(csv_files)} evaluation_results.csv files")
    print("=" * 60)

    # Dictionary to accumulate data for overall average
    all_data = {
        model: {metric: [] for metric in METRICS} for model in MODEL_COLORS.keys()
    }

    series_data = {}
    combined_open_closed_lines = []
    combined_series_avg_lines = []
    scenario_data = {}
    gt_projects = set()

    # Process each subfolder
    for csv_file in sorted(csv_files):
        scenario_name = csv_file.parent.name
        series_name = scenario_name.split("-")[0]

        try:
            # Read the main evaluation results CSV file
            df = pd.read_csv(csv_file)

            # Read retry_summary.csv for error_rate
            retry_file = csv_file.parent / "retry_summary.csv"
            if retry_file.exists():
                retry_df = pd.read_csv(retry_file)
                retry_rate_col = "Retry_Rate(%)"
                if (
                    retry_rate_col not in retry_df.columns
                    and "Error_Rate(%)" in retry_df.columns
                ):
                    retry_rate_col = "Error_Rate(%)"
                retry_df["retry_rate"] = retry_df[retry_rate_col] / 100.0
                # Merge with main dataframe
                df = df.merge(
                    retry_df[["Model", "retry_rate"]],
                    left_on="model",
                    right_on="Model",
                    how="left",
                )
                df = df.drop(columns=["Model"])
                df["retry_rate"] = df["retry_rate"].fillna(0.0)
            else:
                print(f" Warning: retry_summary.csv not found for {scenario_name}")
                df["retry_rate"] = 0.0

            # Read success_rate.csv for success_rate
            success_file = csv_file.parent / "success_rate.csv"
            if success_file.exists():
                success_df = pd.read_csv(success_file)
                # Convert Success_Rate(%) to decimal pass_rate
                success_df["pass_rate"] = success_df["Success_Rate(%)"] / 100.0
                # Merge with main dataframe
                df = df.merge(
                    success_df[["Model", "pass_rate"]],
                    left_on="model",
                    right_on="Model",
                    how="left",
                )
                df = df.drop(columns=["Model"])
                df["pass_rate"] = df["pass_rate"].fillna(0.0)
            else:
                print(f" Warning: success_rate.csv not found for {scenario_name}")
                df["pass_rate"] = 0.0

            df["pass_rate_agg"] = df["pass_rate"]

            pass_rate_label = "pass_rate"
            score_file = csv_file.parent / "score_summary.csv"
            if score_file.exists():
                score_df = pd.read_csv(score_file)
                score_df["average_score"] = score_df["Mean_Score"] / 100.0
                df = df.merge(
                    score_df[["Model", "average_score"]],
                    left_on="model",
                    right_on="Model",
                    how="left",
                )
                df["pass_rate"] = df["average_score"].fillna(df["pass_rate"])
                df = df.drop(columns=["Model", "average_score"])
                pass_rate_label = "average_score"
                gt_projects.add(series_name)

            # Check if all required metrics are present
            missing_metrics = [m for m in METRICS if m not in df.columns]
            if missing_metrics:
                print(f" Skipping {scenario_name}: missing metrics {missing_metrics}")
                continue

            # Store per-scenario data for comparisons
            scenario_data[scenario_name] = {}
            for idx, row in df.iterrows():
                model_name = row["model"]
                scenario_data[scenario_name][model_name] = {
                    metric: row[metric] for metric in METRICS
                }

            # Accumulate data for overall and series-level charts
            for idx, row in df.iterrows():
                model_name = row["model"]
                if model_name in all_data:
                    for metric in METRICS:
                        value = row[metric]
                        all_data[model_name][metric].append(value)
                        series_models = series_data.setdefault(series_name, {})
                        if model_name not in series_models:
                            series_models[model_name] = {m: [] for m in METRICS}
                        series_models[model_name][metric].append(value)

            # Create output path (all PDFs go to RadarCharts, filenames unchanged)
            output_path = output_dir / f"{scenario_name}_radar.pdf"

            # Generate radar chart
            create_radar_chart(
                df,
                scenario_name,
                output_path,
                pass_rate_label=pass_rate_label,
            )

        except Exception as e:
            print(f" Error processing {scenario_name}: {e}")

    print("=" * 60)
    print("All individual radar charts generated successfully!")
    print("=" * 60)

    # Generate overall average chart
    try:
        print("Generating overall average chart...")

        # Calculate averages for each model
        avg_data = []
        for model_name, metrics_data in all_data.items():
            row_data = {"model": model_name}
            for metric in METRICS:
                if metrics_data[metric]:  # Check if there's data
                    row_data[metric] = np.mean(metrics_data[metric])
                else:
                    row_data[metric] = 0
            avg_data.append(row_data)

        # Create DataFrame
        avg_df = pd.DataFrame(avg_data)

        group_rows = []
        for group_name in ["open", "closed"]:
            group_models = [
                m for m, category in MODEL_CATEGORY.items() if category == group_name
            ]
            if not group_models:
                continue
            group_df = avg_df[avg_df["model"].isin(group_models)]
            if group_df.empty:
                continue
            row = {"group": group_name, "model_count": len(group_df)}
            for metric in METRICS:
                row[metric] = float(group_df[metric].mean())
            group_rows.append(row)
        if group_rows:
            header_cols = ["Group", "Model Count"] + METRICS
            header_line = "| " + " | ".join(header_cols) + " |\n"
            separator_line = "| " + " | ".join(["---"] * len(header_cols)) + " |\n"
            md_lines = []
            md_lines.append("# Overall Open vs Closed Models\n\n")
            md_lines.append(header_line)
            md_lines.append(separator_line)
            for row in group_rows:
                row_vals = [
                    row["group"],
                    str(row["model_count"]),
                ]
                for metric in METRICS:
                    val = row[metric]
                    row_vals.append(f"{val:.2f}")
                md_lines.append("| " + " | ".join(row_vals) + " |\n")
            combined_open_closed_lines.extend(md_lines)
            combined_open_closed_lines.append("\n")

        # First, calculate overall summary across all projects
        overall_models_data = {}
        for series_name, models_data in series_data.items():
            for model_name, metrics_data in models_data.items():
                if model_name not in overall_models_data:
                    overall_models_data[model_name] = {m: [] for m in METRICS}
                for metric in METRICS:
                    values = metrics_data.get(metric, [])
                    overall_models_data[model_name][metric].extend(values)

        # Generate overall summary table
        if overall_models_data:
            overall_rows = []
            for model_name, metrics_data in overall_models_data.items():
                row_data = {"model": model_name}
                for metric in METRICS:
                    values = metrics_data.get(metric, [])
                    if values:
                        row_data[metric] = float(np.mean(values))
                    else:
                        row_data[metric] = 0
                overall_rows.append(row_data)

            if overall_rows:
                overall_df = pd.DataFrame(overall_rows)
                combined_series_avg_lines.append("# Overall Summary (All Projects)\n\n")
                combined_series_avg_lines.append(df_to_markdown(overall_df) + "\n\n")

        for series_name, models_data in sorted(series_data.items()):
            series_avg_rows = []
            for model_name, metrics_data in models_data.items():
                row_data = {"model": model_name}
                for metric in METRICS:
                    values = metrics_data.get(metric, [])
                    if values:
                        row_data[metric] = float(np.mean(values))
                    else:
                        row_data[metric] = 0
                series_avg_rows.append(row_data)
            if not series_avg_rows:
                continue
            series_df = pd.DataFrame(series_avg_rows)
            series_pdf_output = output_dir / f"{series_name}_Series_Average_radar.pdf"
            series_pass_rate_label = (
                "average_score" if series_name in gt_projects else "pass_rate"
            )
            create_radar_chart(
                series_df,
                f"{series_name} Series Average",
                series_pdf_output,
                pass_rate_label=series_pass_rate_label,
            )

            combined_series_avg_lines.append(f"# {series_name} Series Average\n\n")
            if series_pass_rate_label != "pass_rate":
                combined_series_avg_lines.append(
                    df_to_markdown(
                        series_df.rename(columns={"pass_rate": series_pass_rate_label})
                    )
                    + "\n\n"
                )
            else:
                combined_series_avg_lines.append(df_to_markdown(series_df) + "\n\n")

            group_rows = []
            for group_name in ["open", "closed"]:
                group_models = [
                    m
                    for m, category in MODEL_CATEGORY.items()
                    if category == group_name
                ]
                if not group_models:
                    continue
                group_df = series_df[series_df["model"].isin(group_models)]
                if group_df.empty:
                    continue
                row = {
                    "series": series_name,
                    "group": group_name,
                    "model_count": len(group_df),
                }
                for metric in METRICS:
                    row[metric] = float(group_df[metric].mean())
                group_rows.append(row)
            if group_rows:
                header_cols = [
                    "Series",
                    "Group",
                    "Model Count",
                ] + series_metric_headers(series_name, gt_projects)
                header_line = "| " + " | ".join(header_cols) + " |\n"
                separator_line = "| " + " | ".join(["---"] * len(header_cols)) + " |\n"
                md_lines = []
                md_lines.append(f"# {series_name} Open vs Closed Models\n\n")
                md_lines.append(header_line)
                md_lines.append(separator_line)
                for row in group_rows:
                    row_vals = [
                        row["series"],
                        row["group"],
                        str(row["model_count"]),
                    ]
                    for metric in METRICS:
                        val = row[metric]
                        row_vals.append(f"{val:.2f}")
                    md_lines.append("| " + " | ".join(row_vals) + " |\n")
                combined_open_closed_lines.extend(md_lines)
                combined_open_closed_lines.append("\n")

        # Combine generated markdown files into a single markdown with titles
        if combined_open_closed_lines:
            combined_md_path = output_dir / "All_Open_vs_Closed_summaries.md"
            combined_md_path.write_text(
                "".join(combined_open_closed_lines),
                encoding="utf-8",
            )
            print(f" Combined markdown created: {combined_md_path}")

        # Combine all Series_Average_radar.csv into a single markdown with titles
        if combined_series_avg_lines:
            combined_csv_md_path = output_dir / "All_Series_Average_tables.md"
            combined_csv_md_path.write_text(
                "".join(combined_series_avg_lines),
                encoding="utf-8",
            )
            print(f" Combined Series Average markdown created: {combined_csv_md_path}")

        # Generate overall chart
        overall_output = output_dir / "Overall_Average_radar.pdf"
        create_radar_chart(avg_df, "Overall Average (All Scenarios)", overall_output)

        legend_output = output_dir / "Model_Legend.pdf"
        create_legend_pdf(legend_output)

        legend_horizontal_output = output_dir / "Model_Legend_horizontal.pdf"
        create_legend_pdf_horizontal(legend_horizontal_output)

        # Generate model comparison markdowns - split into 3 separate files

        # File 1: GPT-5 vs GPT-4o-mini (powerful vs lightweight)
        comparison1_lines = []
        comparison1_lines.append(
            generate_model_pair_comparison(
                "GPT-5",
                "GPT-4o-mini",
                scenario_data,
                "GPT-5 vs GPT-4o-mini (Powerful vs Lightweight)",
                gt_projects=gt_projects,
            )
        )
        comparison1_md_path = output_dir / "Comparison_Powerful_vs_Lightweight.md"
        comparison1_md_path.write_text("".join(comparison1_lines), encoding="utf-8")
        print(f" Powerful vs Lightweight comparison created: {comparison1_md_path}")

        # File 2: Reasoning vs Non-Reasoning models
        comparison2_lines = []
        comparison2_lines.append(
            generate_model_pair_comparison(
                "DeepSeek-R1",
                "DeepSeek-V3-1",
                scenario_data,
                "DeepSeek-R1 vs DeepSeek-V3-1 (Reasoning vs Non-Reasoning)",
                gt_projects=gt_projects,
            )
        )
        comparison2_lines.append("\n---\n\n")
        comparison2_lines.append(
            generate_model_pair_comparison(
                "Gemini-2.5-flash",
                "Gemini-2.5-flash-nothinking",
                scenario_data,
                "Gemini-2.5-flash vs Gemini-2.5-flash-nothinking (Reasoning vs Non-Reasoning)",
                gt_projects=gt_projects,
            )
        )
        comparison2_md_path = output_dir / "Comparison_Reasoning_vs_NonReasoning.md"
        comparison2_md_path.write_text("".join(comparison2_lines), encoding="utf-8")
        print(f" Reasoning vs Non-Reasoning comparison created: {comparison2_md_path}")

        # File 3: MCP vs Hardcoded comparisons
        comparison3_lines = []
        projects = ["MarkdownValidator", "GameBuilder", "EmailResponder"]

        # Add overall summary first
        comparison3_lines.append(
            generate_mcp_overall_comparison(
                projects, scenario_data, gt_projects=gt_projects
            )
        )

        # Then add per-project comparisons
        for project in projects:
            comparison3_lines.append(
                generate_mcp_comparison(project, scenario_data, gt_projects=gt_projects)
            )

        comparison3_md_path = output_dir / "Comparison_MCP_vs_Hardcoded.md"
        comparison3_md_path.write_text("".join(comparison3_lines), encoding="utf-8")
        print(f" MCP vs Hardcoded comparison created: {comparison3_md_path}")

        # File 4: MCP vs A2A comparisons
        comparison4_lines = []
        version_projects = [
            "SQL_assistant",
            "intelligent_recruitment_platform",
            "landing_page_generator",
            "self_evaluation_loop_flow",
            "write_a_book_with_flows",
        ]

        # Add overall summary first
        comparison4_lines.append(
            generate_version_overall_comparison(
                version_projects,
                "-MCP",
                "-A2A",
                scenario_data,
                "MCP",
                "A2A",
                "MCP vs A2A Comparison",
                gt_projects=gt_projects,
            )
        )

        # Then add per-project comparisons
        for project in version_projects:
            comparison4_lines.append(
                generate_version_comparison(
                    project,
                    "-MCP",
                    "-A2A",
                    scenario_data,
                    "MCP",
                    "A2A",
                    gt_projects=gt_projects,
                )
            )

        comparison4_md_path = output_dir / "Comparison_MCP_vs_A2A.md"
        comparison4_md_path.write_text("".join(comparison4_lines), encoding="utf-8")
        print(f" MCP vs A2A comparison created: {comparison4_md_path}")

        # File 5: A2A vs A2A_mix comparisons
        comparison5_lines = []

        # Add overall summary first
        comparison5_lines.append(
            generate_version_overall_comparison(
                version_projects,
                "-A2A",
                "-A2A_mix",
                scenario_data,
                "A2A",
                "A2A_mix",
                "A2A vs A2A_mix Comparison",
                gt_projects=gt_projects,
            )
        )

        # Then add per-project comparisons
        for project in version_projects:
            comparison5_lines.append(
                generate_version_comparison(
                    project,
                    "-A2A",
                    "-A2A_mix",
                    scenario_data,
                    "A2A",
                    "A2A_mix",
                    gt_projects=gt_projects,
                )
            )

        comparison5_md_path = output_dir / "Comparison_A2A_vs_A2A_mix.md"
        comparison5_md_path.write_text("".join(comparison5_lines), encoding="utf-8")
        print(f" A2A vs A2A_mix comparison created: {comparison5_md_path}")

        print("=" * 60)
        print(f" Overall average chart created: {overall_output}")
        print(f" Legend PDF created: {legend_output}")
        print(f" Horizontal legend PDF created: {legend_horizontal_output}")

    except Exception as e:
        print(f" Error generating overall chart: {e}")

    print("=" * 60)
    print("All radar charts completed!")


if __name__ == "__main__":
    main()