visualize_individual_scenarios_v04.py

"""
Individual Scenario Visualization for MVP v0.4.3
3つのシナリオの詳細可視化スクリプト
"""

import json
import matplotlib.pyplot as plt
import numpy as np
from pathlib import Path
import pandas as pd
import seaborn as sns

# 日本語フォント設定（Windows対応 - 警告完全抑制）
import matplotlib
import warnings
import logging
# 全ての警告を抑制
warnings.filterwarnings('ignore')
# matplotlib のログレベルを ERROR に設定
matplotlib.set_loglevel('ERROR')
logging.getLogger('matplotlib').setLevel(logging.ERROR)
logging.getLogger('matplotlib.font_manager').setLevel(logging.ERROR)
# Windows確実対応フォント設定（存在確認不要な汎用設定）
matplotlib.rcParams['font.family'] = ['DejaVu Sans', 'Arial', 'sans-serif']
matplotlib.rcParams['axes.unicode_minus'] = False
plt.rcParams['font.family'] = ['DejaVu Sans', 'Arial', 'sans-serif']
plt.rcParams['axes.unicode_minus'] = False

def visualize_scenario(scenario_name, output_dir, color='#2E86C1'):
    """個別シナリオの詳細可視化"""
    
    # データ読み込み
    history_file = Path(output_dir) / 'training_history.json'
    
    if not history_file.exists():
        print(f"❌ {scenario_name}: データファイルが見つかりません - {history_file}")
        return False
    
    with open(history_file, 'r') as f:
        data = json.load(f)
    
    episode_rewards = data.get('episode_rewards', [])
    episode_costs = data.get('episode_costs', [])
    cost_variances = data.get('episode_cost_variances', [])
    
    if not episode_rewards:
        print(f"❌ {scenario_name}: エピソード報酬データがありません")
        return False
    
    print(f"✅ {scenario_name}: {len(episode_rewards)}エピソードの可視化を開始")
    
    # 2x2のサブプロット作成
    fig, ((ax1, ax2), (ax3, ax4)) = plt.subplots(2, 2, figsize=(16, 12))
    fig.suptitle(f'{scenario_name} Scenario - Detailed Analysis (1000 Episodes)', 
                fontsize=16, fontweight='bold')
    
    episodes = range(1, len(episode_rewards) + 1)
    
    # 1. 学習曲線（移動平均付き）
    ax1.set_title('Learning Curve with Moving Average', fontsize=12, fontweight='bold')
    ax1.plot(episodes, episode_rewards, alpha=0.3, color=color, label='Raw Rewards')
    
    # 移動平均（50エピソード）
    if len(episode_rewards) >= 50:
        window = 50
        moving_avg = np.convolve(episode_rewards, np.ones(window)/window, mode='valid')
        moving_episodes = range(window, len(episode_rewards) + 1)
        ax1.plot(moving_episodes, moving_avg, color=color, linewidth=2, label='Moving Average (50ep)')
    
    ax1.set_xlabel('Episode')
    ax1.set_ylabel('Reward')
    ax1.legend()
    ax1.grid(True, alpha=0.3)
    
    # 2. 収束性分析（最後200エピソード）
    ax2.set_title('Convergence Analysis (Last 200 Episodes)', fontsize=12, fontweight='bold')
    
    if len(episode_rewards) >= 200:
        final_rewards = episode_rewards[-200:]
        final_episodes = range(len(episode_rewards)-199, len(episode_rewards)+1)
        
        ax2.plot(final_episodes, final_rewards, color=color, alpha=0.7)
        
        # 統計情報
        mean_final = np.mean(final_rewards)
        std_final = np.std(final_rewards)
        
        ax2.axhline(y=mean_final, color='red', linestyle='--', alpha=0.8, 
                   label=f'Mean: {mean_final:.1f}')
        ax2.axhline(y=mean_final + std_final, color='orange', linestyle=':', alpha=0.6,
                   label=f'Mean ± Std: {std_final:.1f}')
        ax2.axhline(y=mean_final - std_final, color='orange', linestyle=':', alpha=0.6)
        
        ax2.set_xlabel('Episode')
        ax2.set_ylabel('Reward')
        ax2.legend()
        ax2.grid(True, alpha=0.3)
    else:
        ax2.text(0.5, 0.5, 'Insufficient data\\n(< 200 episodes)', 
                ha='center', va='center', transform=ax2.transAxes)
    
    # 3. Reward VaR (Value at Risk) 分析
    ax3.set_title('Reward VaR (Value at Risk)', fontsize=12, fontweight='bold')
    
    # VaR計算（5%, 10%, 25%パーセンタイル）
    if len(episode_rewards) >= 20:  # 十分なデータがある場合
        # 移動窓でVaRを計算
        window_size = min(100, len(episode_rewards) // 4)
        var_5_percent = []
        var_10_percent = []
        var_25_percent = []
        var_episodes = []
        
        for i in range(window_size, len(episode_rewards) + 1):
            window_rewards = episode_rewards[i-window_size:i]
            var_5_percent.append(np.percentile(window_rewards, 5))
            var_10_percent.append(np.percentile(window_rewards, 10))
            var_25_percent.append(np.percentile(window_rewards, 25))
            var_episodes.append(i)
        
        # VaRプロット
        ax3.plot(var_episodes, var_5_percent, color='red', linewidth=2, alpha=0.8, label='5% VaR')
        ax3.plot(var_episodes, var_10_percent, color='orange', linewidth=2, alpha=0.8, label='10% VaR')
        ax3.plot(var_episodes, var_25_percent, color='green', linewidth=2, alpha=0.8, label='25% VaR')
        
        ax3.set_xlabel('Episode')
        ax3.set_ylabel('VaR (Reward at Risk)')
        ax3.legend()
        ax3.grid(True, alpha=0.3)
        
        # 最終VaR値
        final_var_5 = var_5_percent[-1] if var_5_percent else 0
        final_var_10 = var_10_percent[-1] if var_10_percent else 0
        ax3.text(0.02, 0.98, f'Final 5% VaR: {final_var_5:.2f}\nFinal 10% VaR: {final_var_10:.2f}', 
                transform=ax3.transAxes, va='top', 
                bbox=dict(boxstyle='round', facecolor='lightblue', alpha=0.5))
    else:
        ax3.text(0.5, 0.5, 'Insufficient data for VaR\n(< 20 episodes)', 
                ha='center', va='center', transform=ax3.transAxes)
    
    # 4. 報酬分布分析
    ax4.set_title('Reward Distribution Analysis', fontsize=12, fontweight='bold')
    
    # ヒストグラム
    ax4.hist(episode_rewards, bins=50, alpha=0.7, color=color, density=True)
    
    # 統計情報
    mean_reward = np.mean(episode_rewards)
    median_reward = np.median(episode_rewards)
    std_reward = np.std(episode_rewards)
    
    ax4.axvline(x=mean_reward, color='red', linestyle='--', 
               label=f'Mean: {mean_reward:.1f}')
    ax4.axvline(x=median_reward, color='orange', linestyle='-', 
               label=f'Median: {median_reward:.1f}')
    
    ax4.set_xlabel('Reward')
    ax4.set_ylabel('Density')
    ax4.legend()
    ax4.grid(True, alpha=0.3)
    
    # 統計サマリーをテキストで表示（VaR情報を含む）
    if len(episode_rewards) >= 20:
        final_var_5 = np.percentile(episode_rewards[-100:] if len(episode_rewards) >= 100 else episode_rewards, 5)
        final_var_10 = np.percentile(episode_rewards[-100:] if len(episode_rewards) >= 100 else episode_rewards, 10)
        final_variance = cost_variances[-1] if cost_variances else 0
        
        stats_text = f'''Statistics:
    Episodes: {len(episode_rewards)}
    Mean: {mean_reward:.2f}
    Std: {std_reward:.2f}
    Min: {min(episode_rewards):.2f}
    Max: {max(episode_rewards):.2f}
    5% VaR: {final_var_5:.2f}
    10% VaR: {final_var_10:.2f}'''
    else:
        final_variance = cost_variances[-1] if cost_variances else 0
        stats_text = f'''Statistics:
    Episodes: {len(episode_rewards)}
    Mean: {mean_reward:.2f}
    Std: {std_reward:.2f}
    Min: {min(episode_rewards):.2f}
    Max: {max(episode_rewards):.2f}
    VaR: Insufficient data'''
    
    ax4.text(1.02, 1.0, stats_text, transform=ax4.transAxes, 
            va='top', ha='left', fontsize=9,
            bbox=dict(boxstyle='round', facecolor='lightgray', alpha=0.8))
    
    plt.tight_layout()
    
    # 保存
    save_path = Path(f'scenario_analysis_{scenario_name.lower().replace("-", "_")}_v043.png')
    plt.savefig(save_path, dpi=300, bbox_inches='tight')
    print(f"   📊 グラフ保存: {save_path}")
    
    # 表示
    plt.show()
    
    return True


def main():
    """メイン実行関数"""
    print("=== 3シナリオ個別可視化分析 - MVP v0.4.3 ===")
    print()
    
    # 3つのシナリオ定義
    scenarios = {
        'Safety-First': {
            'output_dir': 'outputs_hvac202_safety_first_enhanced',
            'color': '#FF6B6B',
            'description': '継続運転の安全を最優先とする保全戦略'
        },
        'Cost-Efficient': {
            'output_dir': 'outputs_hvac202_cost_efficient_enhanced',
            'color': '#4ECDC4',
            'description': '予算節約を最優先とする保全戦略'
        },
        'Balanced': {
            'output_dir': 'outputs_hvac202_balanced_enhanced',
            'color': '#45B7D1',
            'description': '安全とコストの最適バランスを目指す保全戦略'
        }
    }
    
    success_count = 0
    
    for scenario_name, config in scenarios.items():
        print(f"🔍 {scenario_name} シナリオ分析中...")
        print(f"   説明: {config['description']}")
        
        success = visualize_scenario(
            scenario_name=scenario_name,
            output_dir=config['output_dir'],
            color=config['color']
        )
        
        if success:
            success_count += 1
        
        print()
    
    print(f"✅ 可視化完了: {success_count}/3 シナリオ")
    print()
    print("生成されたファイル:")
    for scenario_name in scenarios.keys():
        filename = f'scenario_analysis_{scenario_name.lower().replace("-", "_")}_v043.png'
        print(f"  📊 {filename}")


if __name__ == "__main__":
    main()