269 lines
10 KiB
Python
269 lines
10 KiB
Python
"""Shared training loop for value-based VSL agents."""
|
|
from __future__ import annotations
|
|
|
|
import copy
|
|
import inspect
|
|
import os
|
|
from typing import Callable
|
|
|
|
import matplotlib
|
|
import numpy as np
|
|
import yaml
|
|
from tqdm import tqdm
|
|
|
|
from envs.edge_vsl_env import SUMOEdgeVSLEnvironment
|
|
from envs.reward_system import REWARD_COMPONENT_COLUMNS, average_reward_components, init_reward_component_totals
|
|
from utils.config import get_agent_config, get_training_config
|
|
from utils.episode_artifacts import save_training_episode_artifacts
|
|
from utils.logger import TrainingLogger
|
|
from utils.plot import plot_training_curves
|
|
from utils.run_dirs import resolve_run_dirs, write_shared_run_config
|
|
from utils.seeding import derive_seed, resolve_base_seed, set_global_seed
|
|
|
|
matplotlib.use("Agg")
|
|
|
|
|
|
def _build_value_based_agent(agent_builder: Callable[..., object], env, agent_config: dict):
|
|
candidate_kwargs = {
|
|
"state_dim": env.state_dim,
|
|
"num_edges": env.num_controlled_edges,
|
|
"num_actions_per_edge": env.action_dim,
|
|
"hidden_dim": agent_config.get("hidden_dim", 256),
|
|
"mixing_hidden_dim": agent_config.get(
|
|
"mixing_hidden_dim",
|
|
agent_config.get("hidden_dim", 256),
|
|
),
|
|
"learning_rate": agent_config.get("learning_rate", 1e-3),
|
|
"gamma": agent_config.get("gamma", 0.99),
|
|
"epsilon_start": agent_config.get("epsilon_start", 1.0),
|
|
"epsilon_end": agent_config.get("epsilon_end", 0.01),
|
|
"epsilon_decay": agent_config.get("epsilon_decay", 200),
|
|
"buffer_size": agent_config.get("buffer_size", 10000),
|
|
"batch_size": agent_config.get("batch_size", 64),
|
|
"target_update": agent_config.get("target_update", 10),
|
|
"device": agent_config.get("device", "cuda"),
|
|
"edge_feature_dim": env.features_per_edge,
|
|
"time_feature_dim": 3,
|
|
"total_edge_count": env.num_edges,
|
|
"controlled_start_index": env.controlled_edge_start_index,
|
|
"num_corridor_blocks": agent_config.get("num_corridor_blocks", 2),
|
|
"corridor_kernel_size": agent_config.get("corridor_kernel_size", 5),
|
|
"corridor_dropout": agent_config.get("corridor_dropout", 0.05),
|
|
}
|
|
accepted = inspect.signature(agent_builder).parameters
|
|
filtered_kwargs = {
|
|
key: value
|
|
for key, value in candidate_kwargs.items()
|
|
if key in accepted
|
|
}
|
|
return agent_builder(**filtered_kwargs)
|
|
|
|
|
|
def train_sumo_value_based(
|
|
model_key: str,
|
|
model_label: str,
|
|
agent_builder: Callable[..., object],
|
|
*,
|
|
log_dir=None,
|
|
checkpoint_dir=None,
|
|
run_timestamp=None,
|
|
):
|
|
with open("config_sumo_vsl.yaml", "r", encoding="utf-8") as f:
|
|
config = yaml.safe_load(f)
|
|
|
|
agent_config = get_agent_config(config, model_key)
|
|
train_config = get_training_config(config)
|
|
base_seed = resolve_base_seed(train_config)
|
|
set_global_seed(base_seed)
|
|
|
|
_, checkpoint_dir, log_dir = resolve_run_dirs(
|
|
model_key,
|
|
log_dir=log_dir,
|
|
checkpoint_dir=checkpoint_dir,
|
|
run_timestamp=run_timestamp,
|
|
)
|
|
os.makedirs(checkpoint_dir, exist_ok=True)
|
|
os.makedirs(log_dir, exist_ok=True)
|
|
|
|
runtime_config = copy.deepcopy(config)
|
|
runtime_config.setdefault("runtime", {})["output_dir"] = log_dir
|
|
runtime_config["runtime"]["evaluation_mode"] = False
|
|
write_shared_run_config(
|
|
runtime_config,
|
|
log_dir=log_dir,
|
|
checkpoint_dir=checkpoint_dir,
|
|
run_timestamp=run_timestamp,
|
|
)
|
|
|
|
logger = TrainingLogger(log_dir, model_key)
|
|
env = SUMOEdgeVSLEnvironment(runtime_config)
|
|
|
|
state_dim = env.state_dim
|
|
num_edges = env.num_controlled_edges
|
|
num_actions_per_edge = env.action_dim
|
|
|
|
print("=" * 70)
|
|
print(f"{model_label} training - SUMO VSL environment")
|
|
print("=" * 70)
|
|
print(f" State dim: {state_dim}")
|
|
print(f" Controlled edges: {num_edges}")
|
|
print(f" Actions per edge: {num_actions_per_edge}")
|
|
print(f" Episode steps: {env.episode_length}")
|
|
print(f" Control interval: {env.control_interval}s")
|
|
print(f" Hidden dim: {agent_config.get('hidden_dim', 256)}")
|
|
print(f" LR: {agent_config.get('learning_rate', 1e-3)}")
|
|
print(f" Device: {agent_config.get('device', 'cuda')}")
|
|
print()
|
|
|
|
print(f" Global seed: {base_seed if base_seed is not None else 'None (random)'}")
|
|
print()
|
|
|
|
agent = _build_value_based_agent(agent_builder, env, agent_config)
|
|
|
|
num_episodes = train_config["num_episodes"]
|
|
save_freq = train_config.get("save_freq", 50)
|
|
log_freq = train_config.get("log_freq", 10)
|
|
|
|
episode_rewards = []
|
|
episode_throughputs = []
|
|
episode_mean_speeds = []
|
|
episode_speed_variance_norms = []
|
|
episode_ttc_risks = []
|
|
value_losses = []
|
|
best_reward = -float("inf")
|
|
|
|
print("Starting training...\n")
|
|
|
|
try:
|
|
for episode in range(1, num_episodes + 1):
|
|
seed = derive_seed(base_seed, episode)
|
|
state = env.reset(seed=seed)
|
|
episode_reward = 0.0
|
|
episode_throughput = 0.0
|
|
episode_speed = 0.0
|
|
episode_speed_variance_norm = 0.0
|
|
episode_reward_components = init_reward_component_totals()
|
|
episode_ttc_risk = 0.0
|
|
done = False
|
|
step = 0
|
|
|
|
pbar = tqdm(total=env.episode_length, desc=f"Ep {episode}/{num_episodes}", leave=False)
|
|
|
|
while not done:
|
|
action, _, _ = agent.select_action(state, deterministic=False)
|
|
next_state, reward, done, info = env.step(action)
|
|
|
|
agent.store_transition(state, action, reward, next_state, done)
|
|
train_stats = agent.update()
|
|
if train_stats:
|
|
value_losses.append(train_stats["value_loss"])
|
|
|
|
episode_reward += reward
|
|
episode_throughput += info["throughput"]
|
|
episode_speed += info["mean_speed_kmh"]
|
|
episode_speed_variance_norm += info["speed_variance_norm"]
|
|
for column in REWARD_COMPONENT_COLUMNS:
|
|
episode_reward_components[column] += float(info.get(column, 0.0))
|
|
episode_ttc_risk += float(info.get("ttc_risk_rate", 0.0))
|
|
state = next_state
|
|
step += 1
|
|
|
|
pbar.set_postfix(
|
|
r=f"{episode_reward:.1f}",
|
|
tp=f"{info['throughput']:.0f}",
|
|
v=f"{info['mean_speed_kmh']:.1f}",
|
|
)
|
|
pbar.update(1)
|
|
|
|
pbar.close()
|
|
|
|
avg_tp = episode_throughput / max(step, 1)
|
|
avg_speed = episode_speed / max(step, 1)
|
|
avg_speed_variance_norm = episode_speed_variance_norm / max(step, 1)
|
|
avg_reward_components = average_reward_components(episode_reward_components, step)
|
|
|
|
episode_rewards.append(episode_reward)
|
|
episode_throughputs.append(avg_tp)
|
|
episode_mean_speeds.append(avg_speed)
|
|
episode_speed_variance_norms.append(avg_speed_variance_norm)
|
|
episode_ttc_risks.append(episode_ttc_risk)
|
|
|
|
loss_val = np.mean(value_losses[-100:]) if value_losses else None
|
|
logger.log(
|
|
episode,
|
|
episode_reward,
|
|
avg_tp,
|
|
avg_speed,
|
|
speed_variance_norm=avg_speed_variance_norm,
|
|
reward_components=avg_reward_components,
|
|
ttc_risk=episode_ttc_risk,
|
|
value_loss=loss_val,
|
|
)
|
|
|
|
episode_summary = {
|
|
"episode": episode,
|
|
"reward": float(episode_reward),
|
|
"avg_throughput": float(avg_tp),
|
|
"avg_mean_speed_kmh": float(avg_speed),
|
|
"avg_speed_variance_norm": float(avg_speed_variance_norm),
|
|
"ttc_risk": float(episode_ttc_risk),
|
|
"value_loss": float(loss_val) if loss_val is not None else None,
|
|
}
|
|
for column, value in avg_reward_components.items():
|
|
episode_summary[f"avg_{column}"] = float(value)
|
|
save_training_episode_artifacts(
|
|
log_dir=log_dir,
|
|
episode=episode,
|
|
episode_metrics=env.episode_metrics,
|
|
control_edges=env.control_edges,
|
|
summary=episode_summary,
|
|
)
|
|
|
|
if episode_reward > best_reward:
|
|
best_reward = episode_reward
|
|
agent.save(os.path.join(checkpoint_dir, "model_best.pt"))
|
|
|
|
if episode % log_freq == 0:
|
|
recent_rewards = episode_rewards[-log_freq:]
|
|
print(f"\nEpisode {episode}/{num_episodes}")
|
|
print(f" Reward: {episode_reward:.2f} (Avg: {np.mean(recent_rewards):.2f})")
|
|
print(f" Throughput: {avg_tp:.1f} veh/h")
|
|
print(f" Mean Speed: {avg_speed:.1f} km/h")
|
|
print(f" Normalized Speed Variance: {avg_speed_variance_norm:.4f}")
|
|
print(
|
|
" Reward Components: "
|
|
+ ", ".join(
|
|
f"{column}={avg_reward_components[column]:.3f}"
|
|
for column in REWARD_COMPONENT_COLUMNS
|
|
)
|
|
)
|
|
if loss_val is not None:
|
|
print(f" Value Loss: {loss_val:.4f}")
|
|
|
|
if episode % save_freq == 0:
|
|
agent.save(os.path.join(checkpoint_dir, f"model_ep{episode}.pt"))
|
|
|
|
except KeyboardInterrupt:
|
|
print("\nTraining interrupted, saving current model...")
|
|
agent.save(os.path.join(checkpoint_dir, "model_interrupted.pt"))
|
|
finally:
|
|
env.close()
|
|
|
|
agent.save(os.path.join(checkpoint_dir, f"model_ep{num_episodes}.pt"))
|
|
|
|
plot_training_curves(
|
|
episode_rewards,
|
|
episode_throughputs,
|
|
episode_mean_speeds,
|
|
episode_speed_variance_norms,
|
|
episode_ttc_risks,
|
|
save_path=os.path.join(log_dir, "training_curves.png"),
|
|
)
|
|
|
|
print("=" * 70)
|
|
print("Training complete")
|
|
print(f" Best reward: {best_reward:.2f}")
|
|
print(f" Model dir: {checkpoint_dir}")
|
|
print(f" Log dir: {log_dir}")
|
|
print("=" * 70)
|