train_debug.py

#!/usr/bin/env python3
"""
Debug script to test environment and action parsing.
Run this FIRST to verify everything works before training.
"""

import asyncio
import re
from client import JobSchedulerEnvEnv
from models import JobSchedulerEnvAction


def parse_action(text: str) -> str:
    """Extract (job_id, machine_id) from model output."""
    match = re.search(r'\(\s*(\d+)\s*,\s*(\d+)\s*\)', text)
    if match:
        return f"({match.group(1)}, {match.group(2)})"
    return None


async def test_environment():
    """Test if environment works and gives non-constant rewards."""
    print("=" * 60)
    print("Testing Job Scheduler Environment")
    print("=" * 60)

    env = JobSchedulerEnvEnv(base_url="http://localhost:8000")

    # Test 1: Reset
    print("\n1. Testing reset()...")
    try:
        result = await env.reset()
        obs = result.observation
        print(f"   ✓ Reset successful")
        print(f"   - Current time: {obs.current_time}")
        print(f"   - Jobs: {len(obs.job_info)}")
        print(f"   - Machines: {len(obs.machine_info)}")
        print(f"   - Description: {obs.llm_description[:100]}")
    except Exception as e:
        print(f"   ✗ Reset failed: {e}")
        return

    # Test 2: Parse action from description
    print("\n2. Testing action extraction...")
    job_ids = [j["id"] for j in obs.job_info]
    machine_ids = [m["id"] for m in obs.machine_info]

    if job_ids and machine_ids:
        test_action_str = f"({job_ids[0]}, {machine_ids[0]})"
        print(f"   Test action: {test_action_str}")
    else:
        print(f"   ✗ No jobs or machines available")
        return

    # Test 3: Step with valid action
    print("\n3. Testing step() with valid action...")
    try:
        action = JobSchedulerEnvAction(action=test_action_str)
        result = await env.step(action)
        obs = result.observation
        reward = result.reward
        done = result.done

        print(f"   ✓ Step successful")
        print(f"   - Reward: {reward}")
        print(f"   - Done: {done}")
        print(f"   - New description: {obs.llm_description[:100]}")
    except Exception as e:
        print(f"   ✗ Step failed: {e}")
        return

    # Test 4: Step with invalid action
    print("\n4. Testing step() with invalid action...")
    try:
        action = JobSchedulerEnvAction(action="(99999, 99999)")
        result = await env.step(action)
        reward = result.reward
        print(f"   ✓ Invalid action handled")
        print(f"   - Reward for invalid action: {reward}")
    except Exception as e:
        print(f"   ✗ Invalid action caused error: {e}")

    # Test 5: Multiple steps
    print("\n5. Testing multiple episodes...")
    rewards_list = []
    for ep in range(3):
        result = await env.reset()
        obs = result.observation
        ep_reward = 0

        for step in range(5):
            jobs = obs.job_info
            machines = obs.machine_info

            if jobs and machines:
                job_id = jobs[0]["id"]
                machine_id = machines[0]["id"]
                action_str = f"({job_id}, {machine_id})"
                action = JobSchedulerEnvAction(action=action_str)
                result = await env.step(action)
                obs = result.observation
                ep_reward += float(result.reward or 0.0)

                if result.done:
                    break

        rewards_list.append(ep_reward)
        print(f"   Episode {ep + 1}: reward={ep_reward:.2f}")

    avg_reward = sum(rewards_list) / len(rewards_list) if rewards_list else 0
    reward_std = (sum((r - avg_reward) ** 2 for r in rewards_list) / len(rewards_list)) ** 0.5
    print(f"   Average: {avg_reward:.2f}, Std: {reward_std:.2f}")

    if reward_std < 0.1:
        print(f"\n   ⚠️  WARNING: Reward variation is very low (std={reward_std:.4f})")
        print(f"      This will prevent the model from learning!")

    await env.close()
    print("\n" + "=" * 60)
    print("Diagnosis Complete")
    print("=" * 60)


if __name__ == "__main__":
    print("Start the OpenEnv server first:")
    print("  uvicorn Job_Scheduler_Env.server.app:app --reload --port 8000")
    print()

    asyncio.run(test_environment())