td3_bc.py

# source: https://github.com/sfujim/TD3_BC
# https://arxiv.org/pdf/2106.06860.pdf
import copy
import os
import random
import uuid
from dataclasses import asdict, dataclass
from pathlib import Path
from typing import Any, Dict, List, Optional, Tuple, Union

import d4rl
import gym
import numpy as np
import pyrallis
import torch
import torch.nn as nn
import torch.nn.functional as F
import wandb

TensorBatch = List[torch.Tensor]


@dataclass
class TrainConfig:
    # wandb project name
    project: str = "CORL"
    # wandb group name
    group: str = "TD3_BC-D4RL"
    # wandb run name
    name: str = "TD3_BC"
    # training dataset and evaluation environment
    env: str = "halfcheetah-medium-expert-v2"
    # coefficient for the Q-function in actor loss
    alpha: float = 2.5
    # discount factor
    discount: float = 0.99
    # standard deviation for the gaussian exploration noise
    expl_noise: float = 0.1
    # coefficient for the target critic Polyak's update
    tau: float = 0.005
    # scalig coefficient for the noise added to
    # target actor during critic update
    policy_noise: float = 0.2
    # range for the target actor noise clipping
    noise_clip: float = 0.5
    # actor update delay
    policy_freq: int = 2
    # total gradient updates during training
    max_timesteps: int = int(1e6)
    # maximum size of the replay buffer
    buffer_size: int = 2_000_000
    # training batch size
    batch_size: int = 256
    # whether to normalize states
    normalize: bool = True
    # whether to normalize reward (like in IQL)
    normalize_reward: bool = False
    # evaluation frequency, will evaluate every eval_freq training steps
    eval_freq: int = int(5e3)
    # number of episodes to run during evaluation
    n_episodes: int = 10
    # path for checkpoints saving, optional
    checkpoints_path: Optional[str] = None
    # file name for loading a model, optional
    load_model: str = ""
    # training random seed
    seed: int = 0
    # training device
    device: str = "cuda"

    def __post_init__(self):
        self.name = f"{self.name}-{self.env}-{str(uuid.uuid4())[:8]}"
        if self.checkpoints_path is not None:
            self.checkpoints_path = os.path.join(self.checkpoints_path, self.name)


def soft_update(target: nn.Module, source: nn.Module, tau: float):
    for target_param, source_param in zip(target.parameters(), source.parameters()):
        target_param.data.copy_((1 - tau) * target_param.data + tau * source_param.data)


def compute_mean_std(states: np.ndarray, eps: float) -> Tuple[np.ndarray, np.ndarray]:
    mean = states.mean(0)
    std = states.std(0) + eps
    return mean, std


def normalize_states(states: np.ndarray, mean: np.ndarray, std: np.ndarray):
    return (states - mean) / std


def wrap_env(
    env: gym.Env,
    state_mean: Union[np.ndarray, float] = 0.0,
    state_std: Union[np.ndarray, float] = 1.0,
    reward_scale: float = 1.0,
) -> gym.Env:
    # PEP 8: E731 do not assign a lambda expression, use a def
    def normalize_state(state):
        return (
            state - state_mean
        ) / state_std  # epsilon should be already added in std.

    def scale_reward(reward):
        # Please be careful, here reward is multiplied by scale!
        return reward_scale * reward

    env = gym.wrappers.TransformObservation(env, normalize_state)
    if reward_scale != 1.0:
        env = gym.wrappers.TransformReward(env, scale_reward)
    return env


class ReplayBuffer:
    def __init__(
        self,
        state_dim: int,
        action_dim: int,
        buffer_size: int,
        device: str = "cpu",
    ):
        self._buffer_size = buffer_size
        self._pointer = 0
        self._size = 0

        self._states = torch.zeros(
            (buffer_size, state_dim), dtype=torch.float32, device=device
        )
        self._actions = torch.zeros(
            (buffer_size, action_dim), dtype=torch.float32, device=device
        )
        self._rewards = torch.zeros((buffer_size, 1), dtype=torch.float32, device=device)
        self._next_states = torch.zeros(
            (buffer_size, state_dim), dtype=torch.float32, device=device
        )
        self._dones = torch.zeros((buffer_size, 1), dtype=torch.float32, device=device)
        self._device = device

    def _to_tensor(self, data: np.ndarray) -> torch.Tensor:
        return torch.tensor(data, dtype=torch.float32, device=self._device)

    # Loads data in d4rl format, i.e. from Dict[str, np.array].
    def load_d4rl_dataset(self, data: Dict[str, np.ndarray]):
        if self._size != 0:
            raise ValueError("Trying to load data into non-empty replay buffer")
        n_transitions = data["observations"].shape[0]
        if n_transitions > self._buffer_size:
            raise ValueError(
                "Replay buffer is smaller than the dataset you are trying to load!"
            )
        self._states[:n_transitions] = self._to_tensor(data["observations"])
        self._actions[:n_transitions] = self._to_tensor(data["actions"])
        self._rewards[:n_transitions] = self._to_tensor(data["rewards"][..., None])
        self._next_states[:n_transitions] = self._to_tensor(data["next_observations"])
        self._dones[:n_transitions] = self._to_tensor(data["terminals"][..., None])
        self._size += n_transitions
        self._pointer = min(self._size, n_transitions)

        print(f"Dataset size: {n_transitions}")

    def sample(self, batch_size: int) -> TensorBatch:
        indices = np.random.randint(0, min(self._size, self._pointer), size=batch_size)
        states = self._states[indices]
        actions = self._actions[indices]
        rewards = self._rewards[indices]
        next_states = self._next_states[indices]
        dones = self._dones[indices]
        return [states, actions, rewards, next_states, dones]

    def add_transition(self):
        # Use this method to add new data into the replay buffer during fine-tuning.
        # I left it unimplemented since now we do not do fine-tuning.
        raise NotImplementedError


def set_seed(
    seed: int, env: Optional[gym.Env] = None, deterministic_torch: bool = False
):
    if env is not None:
        env.seed(seed)
        env.action_space.seed(seed)
    os.environ["PYTHONHASHSEED"] = str(seed)
    np.random.seed(seed)
    random.seed(seed)
    torch.manual_seed(seed)
    torch.use_deterministic_algorithms(deterministic_torch)


def wandb_init(config: dict) -> None:
    wandb.init(
        config=config,
        project=config["project"],
        group=config["group"],
        name=config["name"],
        id=str(uuid.uuid4()),
    )
    wandb.run.save()


@torch.no_grad()
def eval_actor(
    env: gym.Env, actor: nn.Module, device: str, n_episodes: int, seed: int
) -> np.ndarray:
    env.seed(seed)
    actor.eval()
    episode_rewards = []
    for _ in range(n_episodes):
        state, done = env.reset(), False
        episode_reward = 0.0
        while not done:
            action = actor.act(state, device)
            state, reward, done, _ = env.step(action)
            episode_reward += reward
        episode_rewards.append(episode_reward)

    actor.train()
    return np.asarray(episode_rewards)


def return_reward_range(dataset, max_episode_steps):
    returns, lengths = [], []
    ep_ret, ep_len = 0.0, 0
    for r, d in zip(dataset["rewards"], dataset["terminals"]):
        ep_ret += float(r)
        ep_len += 1
        if d or ep_len == max_episode_steps:
            returns.append(ep_ret)
            lengths.append(ep_len)
            ep_ret, ep_len = 0.0, 0
    lengths.append(ep_len)  # but still keep track of number of steps
    assert sum(lengths) == len(dataset["rewards"])
    return min(returns), max(returns)


def modify_reward(dataset, env_name, max_episode_steps=1000):
    if any(s in env_name for s in ("halfcheetah", "hopper", "walker2d")):
        min_ret, max_ret = return_reward_range(dataset, max_episode_steps)
        dataset["rewards"] /= max_ret - min_ret
        dataset["rewards"] *= max_episode_steps
    elif "antmaze" in env_name:
        dataset["rewards"] -= 1.0


class Actor(nn.Module):
    def __init__(self, state_dim: int, action_dim: int, max_action: float):
        super(Actor, self).__init__()

        self.net = nn.Sequential(
            nn.Linear(state_dim, 256),
            nn.ReLU(),
            nn.Linear(256, 256),
            nn.ReLU(),
            nn.Linear(256, action_dim),
            nn.Tanh(),
        )

        self.max_action = max_action

    def forward(self, state: torch.Tensor) -> torch.Tensor:
        return self.max_action * self.net(state)

    @torch.no_grad()
    def act(self, state: np.ndarray, device: str = "cpu") -> np.ndarray:
        state = torch.tensor(state.reshape(1, -1), device=device, dtype=torch.float32)
        return self(state).cpu().data.numpy().flatten()


class Critic(nn.Module):
    def __init__(self, state_dim: int, action_dim: int):
        super(Critic, self).__init__()

        self.net = nn.Sequential(
            nn.Linear(state_dim + action_dim, 256),
            nn.ReLU(),
            nn.Linear(256, 256),
            nn.ReLU(),
            nn.Linear(256, 1),
        )

    def forward(self, state: torch.Tensor, action: torch.Tensor) -> torch.Tensor:
        sa = torch.cat([state, action], 1)
        return self.net(sa)


class TD3_BC:
    def __init__(
        self,
        max_action: float,
        actor: nn.Module,
        actor_optimizer: torch.optim.Optimizer,
        critic_1: nn.Module,
        critic_1_optimizer: torch.optim.Optimizer,
        critic_2: nn.Module,
        critic_2_optimizer: torch.optim.Optimizer,
        discount: float = 0.99,
        tau: float = 0.005,
        policy_noise: float = 0.2,
        noise_clip: float = 0.5,
        policy_freq: int = 2,
        alpha: float = 2.5,
        device: str = "cpu",
    ):
        self.actor = actor
        self.actor_target = copy.deepcopy(actor)
        self.actor_optimizer = actor_optimizer
        self.critic_1 = critic_1
        self.critic_1_target = copy.deepcopy(critic_1)
        self.critic_1_optimizer = critic_1_optimizer
        self.critic_2 = critic_2
        self.critic_2_target = copy.deepcopy(critic_2)
        self.critic_2_optimizer = critic_2_optimizer

        self.max_action = max_action
        self.discount = discount
        self.tau = tau
        self.policy_noise = policy_noise
        self.noise_clip = noise_clip
        self.policy_freq = policy_freq
        self.alpha = alpha

        self.total_it = 0
        self.device = device

    def train(self, batch: TensorBatch) -> Dict[str, float]:
        log_dict = {}
        self.total_it += 1

        state, action, reward, next_state, done = batch
        not_done = 1 - done

        with torch.no_grad():
            # Select action according to actor and add clipped noise
            noise = (torch.randn_like(action) * self.policy_noise).clamp(
                -self.noise_clip, self.noise_clip
            )

            next_action = (self.actor_target(next_state) + noise).clamp(
                -self.max_action, self.max_action
            )

            # Compute the target Q value
            target_q1 = self.critic_1_target(next_state, next_action)
            target_q2 = self.critic_2_target(next_state, next_action)
            target_q = torch.min(target_q1, target_q2)
            target_q = reward + not_done * self.discount * target_q

        # Get current Q estimates
        current_q1 = self.critic_1(state, action)
        current_q2 = self.critic_2(state, action)

        # Compute critic loss
        critic_loss = F.mse_loss(current_q1, target_q) + F.mse_loss(current_q2, target_q)
        log_dict["critic_loss"] = critic_loss.item()
        # Optimize the critic
        self.critic_1_optimizer.zero_grad()
        self.critic_2_optimizer.zero_grad()
        critic_loss.backward()
        self.critic_1_optimizer.step()
        self.critic_2_optimizer.step()

        # Delayed actor updates
        if self.total_it % self.policy_freq == 0:
            # Compute actor loss
            pi = self.actor(state)
            q = self.critic_1(state, pi)
            lmbda = self.alpha / q.abs().mean().detach()

            actor_loss = -lmbda * q.mean() + F.mse_loss(pi, action)
            log_dict["actor_loss"] = actor_loss.item()
            # Optimize the actor
            self.actor_optimizer.zero_grad()
            actor_loss.backward()
            self.actor_optimizer.step()

            # Update the frozen target models
            soft_update(self.critic_1_target, self.critic_1, self.tau)
            soft_update(self.critic_2_target, self.critic_2, self.tau)
            soft_update(self.actor_target, self.actor, self.tau)

        return log_dict

    def state_dict(self) -> Dict[str, Any]:
        return {
            "critic_1": self.critic_1.state_dict(),
            "critic_1_optimizer": self.critic_1_optimizer.state_dict(),
            "critic_2": self.critic_2.state_dict(),
            "critic_2_optimizer": self.critic_2_optimizer.state_dict(),
            "actor": self.actor.state_dict(),
            "actor_optimizer": self.actor_optimizer.state_dict(),
            "total_it": self.total_it,
        }

    def load_state_dict(self, state_dict: Dict[str, Any]):
        self.critic_1.load_state_dict(state_dict["critic_1"])
        self.critic_1_optimizer.load_state_dict(state_dict["critic_1_optimizer"])
        self.critic_1_target = copy.deepcopy(self.critic_1)

        self.critic_2.load_state_dict(state_dict["critic_2"])
        self.critic_2_optimizer.load_state_dict(state_dict["critic_2_optimizer"])
        self.critic_2_target = copy.deepcopy(self.critic_2)

        self.actor.load_state_dict(state_dict["actor"])
        self.actor_optimizer.load_state_dict(state_dict["actor_optimizer"])
        self.actor_target = copy.deepcopy(self.actor)

        self.total_it = state_dict["total_it"]


@pyrallis.wrap()
def train(config: TrainConfig):
    env = gym.make(config.env)

    state_dim = env.observation_space.shape[0]
    action_dim = env.action_space.shape[0]

    dataset = d4rl.qlearning_dataset(env)

    if config.normalize_reward:
        modify_reward(dataset, config.env)

    if config.normalize:
        state_mean, state_std = compute_mean_std(dataset["observations"], eps=1e-3)
    else:
        state_mean, state_std = 0, 1

    dataset["observations"] = normalize_states(
        dataset["observations"], state_mean, state_std
    )
    dataset["next_observations"] = normalize_states(
        dataset["next_observations"], state_mean, state_std
    )
    env = wrap_env(env, state_mean=state_mean, state_std=state_std)
    replay_buffer = ReplayBuffer(
        state_dim,
        action_dim,
        config.buffer_size,
        config.device,
    )
    replay_buffer.load_d4rl_dataset(dataset)

    max_action = float(env.action_space.high[0])

    if config.checkpoints_path is not None:
        print(f"Checkpoints path: {config.checkpoints_path}")
        os.makedirs(config.checkpoints_path, exist_ok=True)
        with open(os.path.join(config.checkpoints_path, "config.yaml"), "w") as f:
            pyrallis.dump(config, f)

    # Set seeds
    seed = config.seed
    set_seed(seed, env)

    actor = Actor(state_dim, action_dim, max_action).to(config.device)
    actor_optimizer = torch.optim.Adam(actor.parameters(), lr=3e-4)

    critic_1 = Critic(state_dim, action_dim).to(config.device)
    critic_1_optimizer = torch.optim.Adam(critic_1.parameters(), lr=3e-4)
    critic_2 = Critic(state_dim, action_dim).to(config.device)
    critic_2_optimizer = torch.optim.Adam(critic_2.parameters(), lr=3e-4)

    kwargs = {
        "max_action": max_action,
        "actor": actor,
        "actor_optimizer": actor_optimizer,
        "critic_1": critic_1,
        "critic_1_optimizer": critic_1_optimizer,
        "critic_2": critic_2,
        "critic_2_optimizer": critic_2_optimizer,
        "discount": config.discount,
        "tau": config.tau,
        "device": config.device,
        # TD3
        "policy_noise": config.policy_noise * max_action,
        "noise_clip": config.noise_clip * max_action,
        "policy_freq": config.policy_freq,
        # TD3 + BC
        "alpha": config.alpha,
    }

    print("---------------------------------------")
    print(f"Training TD3 + BC, Env: {config.env}, Seed: {seed}")
    print("---------------------------------------")

    # Initialize actor
    trainer = TD3_BC(**kwargs)

    if config.load_model != "":
        policy_file = Path(config.load_model)
        trainer.load_state_dict(torch.load(policy_file))
        actor = trainer.actor

    wandb_init(asdict(config))

    evaluations = []
    for t in range(int(config.max_timesteps)):
        batch = replay_buffer.sample(config.batch_size)
        batch = [b.to(config.device) for b in batch]
        log_dict = trainer.train(batch)
        wandb.log(log_dict, step=trainer.total_it)
        # Evaluate episode
        if (t + 1) % config.eval_freq == 0:
            print(f"Time steps: {t + 1}")
            eval_scores = eval_actor(
                env,
                actor,
                device=config.device,
                n_episodes=config.n_episodes,
                seed=config.seed,
            )
            eval_score = eval_scores.mean()
            normalized_eval_score = env.get_normalized_score(eval_score) * 100.0
            evaluations.append(normalized_eval_score)
            print("---------------------------------------")
            print(
                f"Evaluation over {config.n_episodes} episodes: "
                f"{eval_score:.3f} , D4RL score: {normalized_eval_score:.3f}"
            )
            print("---------------------------------------")

            if config.checkpoints_path is not None:
                torch.save(
                    trainer.state_dict(),
                    os.path.join(config.checkpoints_path, f"checkpoint_{t}.pt"),
                )

            wandb.log(
                {"d4rl_normalized_score": normalized_eval_score},
                step=trainer.total_it,
            )


if __name__ == "__main__":
    train()