Spaces:
Sleeping
Sleeping
File size: 9,320 Bytes
b00d5d5 | 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 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 | """
Trainer — manages the training loop for any BaseAgent.
Features:
• Per-episode logging (episode, reward, waiting time, queue, throughput)
• Automatic best-model saving
• Periodic checkpoints
• Early stopping
• DQN target-network updates
• Graceful error recovery (episode-level try/except)
• Optional tqdm progress bar
"""
from __future__ import annotations
import sys
import traceback
from pathlib import Path
import numpy as np
try:
from tqdm import tqdm
_TQDM = True
except ImportError:
_TQDM = False
from utils.logger import setup_logger
from utils.metrics import MetricsTracker
class Trainer:
"""
Orchestrates the RL training loop.
Args:
env: A Gymnasium-compatible environment.
agent: Any agent that inherits :class:`BaseAgent`.
config: The project config module.
"""
def __init__(self, env, agent, config):
self.env = env
self.agent = agent
self.config = config
# Directories
config.RESULTS_DIR.mkdir(parents=True, exist_ok=True)
config.MODELS_DIR.mkdir(parents=True, exist_ok=True)
(config.RESULTS_DIR / "logs").mkdir(exist_ok=True)
(config.RESULTS_DIR / "plots").mkdir(exist_ok=True)
(config.RESULTS_DIR / "checkpoints").mkdir(exist_ok=True)
# Logger
log_file = config.RESULTS_DIR / "logs" / "training.log"
self.logger = setup_logger("trainer", log_file=str(log_file))
# Metrics
self.metrics = MetricsTracker()
# State
self.best_reward: float = -np.inf
self.episodes_without_improvement: int = 0
self.current_episode: int = 0
self.total_steps: int = 0
self.logger.info("=" * 70)
self.logger.info("TRAINER READY")
self.logger.info(f" Agent type : {config.AGENT_TYPE}")
self.logger.info(f" Results dir: {config.RESULTS_DIR}")
self.logger.info("=" * 70)
# ------------------------------------------------------------------
# Public
# ------------------------------------------------------------------
def train(self, num_episodes: int):
"""
Run the training loop for *num_episodes* episodes.
Args:
num_episodes: Number of episodes to train.
"""
self.logger.info(f"Starting training — {num_episodes} episodes")
iterator = (
tqdm(range(1, num_episodes + 1), desc="Training", unit="ep")
if _TQDM
else range(1, num_episodes + 1)
)
try:
for episode in iterator:
self.current_episode = episode
try:
ep_reward, ep_info = self._run_episode(training=True)
except KeyboardInterrupt:
self.logger.info("Training interrupted by user.")
self._save_checkpoint(episode, emergency=True)
raise
except Exception as exc:
self.logger.error(f"Episode {episode} error: {exc}")
self.logger.debug(traceback.format_exc())
continue
# Record metrics
self.metrics.add("episode_reward", ep_reward)
self.metrics.add("average_waiting_time",
ep_info.get("average_waiting_time", 0.0))
self.metrics.add("average_queue_length",
ep_info.get("total_queue_length", 0.0))
self.metrics.add("throughput",
ep_info.get("vehicles_passed", 0))
# Per-episode log
self.logger.info(
f"Ep {episode:4d}/{num_episodes} "
f"reward={ep_reward:8.2f} "
f"wait={ep_info.get('average_waiting_time', 0):7.1f} "
f"queue={ep_info.get('total_queue_length', 0):6.1f} "
f"thru={ep_info.get('vehicles_passed', 0):4d}"
)
# DQN: sync target network
if hasattr(self.agent, "update_target_network"):
freq = self.config.DQN_CONFIG.get("target_update", 10)
if episode % freq == 0:
self.agent.update_target_network()
# Save best model
if ep_reward > self.best_reward:
self.best_reward = ep_reward
self.episodes_without_improvement = 0
self._save_best_model(episode, ep_reward)
else:
self.episodes_without_improvement += 1
# Periodic checkpoint
if episode % self.config.SAVE_FREQUENCY == 0:
self._save_checkpoint(episode)
# Periodic summary
if episode % 100 == 0:
self._log_summary(episode, num_episodes)
# Early stopping
if self.episodes_without_improvement >= self.config.EARLY_STOPPING_PATIENCE:
self.logger.info(
f"Early stopping at episode {episode} "
f"(no improvement for "
f"{self.config.EARLY_STOPPING_PATIENCE} episodes)."
)
break
except KeyboardInterrupt:
self.logger.info("Exiting gracefully.")
sys.exit(0)
self.logger.info("=" * 70)
self.logger.info("TRAINING COMPLETE")
self._log_final_summary()
self._save_metrics()
self._plot_results()
# ------------------------------------------------------------------
# Internal
# ------------------------------------------------------------------
def _run_episode(self, training: bool = True) -> tuple[float, dict]:
"""Execute one full episode."""
state, _ = self.env.reset()
ep_reward = 0.0
done = False
info: dict = {}
max_steps = self.config.EPISODE_LENGTH * 2
steps = 0
while not done and steps < max_steps:
action = self.agent.select_action(state, training=training)
next_state, reward, terminated, truncated, info = self.env.step(action)
done = terminated or truncated
if training:
loss = self.agent.train_step(state, action, reward, next_state, done)
if loss is not None:
self.metrics.add("loss", float(loss))
state = next_state
ep_reward += reward
steps += 1
self.total_steps += 1
return ep_reward, info
def _save_best_model(self, episode: int, reward: float):
path = self.config.MODELS_DIR / f"{self.config.AGENT_TYPE}_best.pth"
try:
self.agent.save(str(path))
self.logger.info(
f"[OK] Best model saved reward={reward:.2f} (episode {episode})"
)
except Exception as exc:
self.logger.error(f"[FAIL] Could not save best model: {exc}")
def _save_checkpoint(self, episode: int, emergency: bool = False):
tag = "emergency" if emergency else f"ep{episode}"
path = (
self.config.RESULTS_DIR
/ "checkpoints"
/ f"{self.config.AGENT_TYPE}_{tag}.pth"
)
try:
self.agent.save(str(path))
self.logger.info(f"[OK] Checkpoint saved -> {path}")
except Exception as exc:
self.logger.error(f"[FAIL] Could not save checkpoint: {exc}")
def _log_summary(self, episode: int, total: int):
n = min(100, episode)
self.logger.info("-" * 70)
self.logger.info(f"Summary ep {episode}/{total}")
self.logger.info(
f" Avg reward (last {n}): "
f"{self.metrics.get_mean('episode_reward', last_n=n):8.2f}"
)
self.logger.info(
f" Avg wait (last {n}): "
f"{self.metrics.get_mean('average_waiting_time', last_n=n):8.2f}"
)
self.logger.info(f" Best reward so far : {self.best_reward:8.2f}")
self.logger.info("-" * 70)
def _log_final_summary(self):
all_r = self.metrics.get("episode_reward")
if not all_r:
return
self.logger.info("FINAL STATISTICS")
self.logger.info(f" Total episodes : {len(all_r)}")
self.logger.info(f" Best reward : {self.best_reward:.2f}")
self.logger.info(f" Mean reward : {np.mean(all_r):.2f}")
self.logger.info(f" Std reward : {np.std(all_r):.2f}")
def _save_metrics(self):
path = self.config.RESULTS_DIR / "metrics.json"
try:
self.metrics.save(path)
self.logger.info(f"[OK] Metrics saved -> {path}")
except Exception as exc:
self.logger.warning(f"Could not save metrics: {exc}")
def _plot_results(self):
try:
from utils.visualizer import plot_training_curves
save = self.config.RESULTS_DIR / "plots" / f"{self.config.AGENT_TYPE}_training.png"
plot_training_curves(self.metrics, save_path=save)
except Exception as exc:
self.logger.warning(f"Could not plot results: {exc}")
|