File size: 2,356 Bytes
ef737d3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 | import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import numpy as np
def generate_winning_plots(rewards_per_episode, losses, baseline_score, trained_scores):
"""
Generates the premium, judge-ready plots for the Autonomy Calibration Benchmark.
"""
# 1. Reward Curve (Calibration Accuracy)
plt.figure(figsize=(10, 5))
plt.plot(rewards_per_episode, color='#27AE60', linewidth=2.5, alpha=0.3)
# Smoothed trend
smooth_rewards = np.convolve(rewards_per_episode, np.ones(10)/10, mode='valid')
plt.plot(range(9, len(rewards_per_episode)), smooth_rewards, color='#1A8A4A', linewidth=3, label='Calibrated Policy Reward')
plt.axhline(y=baseline_score, color='#E74C3C', linestyle='--', linewidth=2, label=f'Rule-Based Baseline ({baseline_score})')
plt.title('🛡️ Autonomy Calibration: Training Progress', fontsize=14, fontweight='bold')
plt.xlabel('Training Episode')
plt.ylabel('Episode Reward (0.01 - 0.99)')
plt.legend()
plt.grid(True, alpha=0.2)
plt.tight_layout()
plt.savefig('reward_curve.png', dpi=150)
plt.close()
# 2. Policy Loss (Divergence)
plt.figure(figsize=(10, 5))
plt.plot(losses, color='#2980B9', linewidth=2)
plt.title('📈 GRPOTrainer Policy Loss', fontsize=14, fontweight='bold')
plt.xlabel('Training Step')
plt.ylabel('Loss')
plt.grid(True, alpha=0.2)
plt.tight_layout()
plt.savefig('loss_curve.png', dpi=150)
plt.close()
# 3. Final Comparison (The ROI)
plt.figure(figsize=(8, 6))
categories = ['Rule-Based Baseline', 'Trained Agent (GRPO)']
values = [baseline_score, np.mean(trained_scores)]
bars = plt.bar(categories, values, color=['#BDC3C7', '#2ECC71'], width=0.6)
plt.title('🏆 Performance Uplift: Accuracy + Calibration', fontsize=14, fontweight='bold')
plt.ylabel('Average Episode Reward')
plt.ylim(0, 1.0)
# Add value labels
for bar in bars:
height = bar.get_height()
plt.text(bar.get_x() + bar.get_width()/2., height + 0.02,
f'{height:.3f}', ha='center', va='bottom', fontsize=12, fontweight='bold')
plt.tight_layout()
plt.savefig('baseline_vs_trained.png', dpi=150)
plt.close()
print("✅ Winning visuals generated: reward_curve.png, loss_curve.png, baseline_vs_trained.png")
|