ddpg_agent.py

import numpy as np
import random
import copy
from collections import namedtuple, deque

from model import Actor, Critic

import torch
import torch.nn.functional as F
import torch.optim as optim



class Agent():
    """Interacts with and learns from the environment."""
    
    def __init__(self, index, config):
        """Initialize an Agent object.
        
        Params
        ======
            state_size (int): dimension of each state
            action_size (int): dimension of each action
            random_seed (int): random seed
        """
        self.index = index
        self.config = config
        self.seed = random.seed(config.general.seed)
        self.device = config.general.device

        # Actor Network (w/ Target Network)
        self.actor_local = Actor(config).to(self.device)
        self.actor_target = Actor(config).to(self.device)
        self.actor_optimizer = optim.Adam(self.actor_local.parameters(), lr=config.actor.lr)

        # Critic Network (w/ Target Network)
        self.critic_local = Critic(config).to(self.device)
        self.critic_target = Critic(config).to(self.device)
        self.critic_optimizer = optim.Adam(self.critic_local.parameters(), lr=config.critic.lr, 
                                           weight_decay=config.hp.weight_decay)

        # Noise process
        self.noise = OUNoise(config.env.action_size, self.seed)

        # Replay memory
        self.memory = ReplayBuffer(config)
        
        # Update Frequency
        self.t_step = 0
        self.update_every = config.hp.update_every
        self.tau = config.hp.tau
    
    def step(self, states, actions, rewards, next_states, dones):
        """Save experience in replay memory, and use random sample from buffer to learn."""
        # Save experience / reward
        # self.memory.add(state, action, reward, next_state, done)
        for state, action, reward, next_state, done in zip(states, actions, rewards, next_states, dones):
            self.memory.add(state, action, reward, next_state, done)
        
        self.t_step = (self.t_step + 1)%self.update_every
        if self.t_step == 0:
            # Learn, if enough samples are available in memory
            if len(self.memory) > BATCH_SIZE:
                experiences = self.memory.sample()
                self.learn(experiences, GAMMA)

    def act(self, state, add_noise=True):
        """Returns actions for given state as per current policy."""
        state = torch.from_numpy(state).float().to(self.device)
        self.actor_local.eval()
        with torch.no_grad():
            action = self.actor_local(state.reshape(1, -1)).cpu().data.numpy()
        self.actor_local.train()
        if add_noise:
            action += self.noise.sample()
        return np.clip(action, -1, 1)

    def reset(self):
        self.noise.reset()

    def learn(self, agent_id, experiences, gamma, all_next_actions, all_actions):
        """Update policy and value parameters using given batch of experience tuples.
        Params
        ======
            experiences (Tuple[torch.Tensor]): tuple of (s, a, r, s', done) tuples
            gamma (float): discount factor
            all_next_actions (list): each agent's next_action (as calculated by it's actor)
            all_actions (list): each agent's action (as calculated by it's actor)
        """

        states, actions, rewards, next_states, dones = experiences

        # ---------------------------- update critic ---------------------------- #
        # get predicted next-state actions and Q values from target models
        self.critic_optimizer.zero_grad()
        agent_id = torch.tensor([agent_id]).to(self.device)
        actions_next = torch.cat(all_next_actions, dim=1).to(self.device)
        with torch.no_grad():
            q_targets_next = self.critic_target(next_states, actions_next)
        # compute Q targets for current states (y_i)
        q_expected = self.critic_local(states, actions)
        # q_targets = reward of this timestep + discount * Q(st+1,at+1) from target network
        q_targets = rewards.index_select(1, agent_id) + (gamma * q_targets_next * (1 - dones.index_select(1, agent_id)))
        # compute critic loss
        critic_loss = F.mse_loss(q_expected, q_targets.detach())
        self.critic_loss = critic_loss.item()  # for tensorboard logging
        # minimize loss
        critic_loss.backward()
        torch.nn.utils.clip_grad_norm_(self.critic_local.parameters(), 1)
        self.critic_optimizer.step()

        # ---------------------------- update actor ---------------------------- #
        # compute actor loss
        self.actor_optimizer.zero_grad()
        # detach actions from other agents
        actions_pred = [actions if i == self.index else actions.detach() for i, actions in enumerate(all_actions)]
        actions_pred = torch.cat(actions_pred, dim=1).to(self.device)
        actor_loss = -self.critic_local(states, actions_pred).mean()
        self.actor_loss = actor_loss.item()  # calculate policy gradient
        # minimize loss
        actor_loss.backward()
        self.actor_optimizer.step()

        # ----------------------- update target networks ----------------------- #
        self.soft_update(self.critic_local, self.critic_target, self.tau)
        self.soft_update(self.actor_local, self.actor_target, self.tau)

    def soft_update(self, local_model, target_model, tau):
        """Soft update model parameters.
        θ_target = τ*θ_local + (1 - τ)*θ_target

        Params
        ======
            local_model: PyTorch model (weights will be copied from)
            target_model: PyTorch model (weights will be copied to)
            tau (float): interpolation parameter 
        """
        for target_param, local_param in zip(target_model.parameters(), local_model.parameters()):
            target_param.data.copy_(tau*local_param.data + (1.0-tau)*target_param.data)

class OUNoise:
    """Ornstein-Uhlenbeck process."""

    def __init__(self, size, seed, mu=0., theta=0.15, sigma=0.2):
        """Initialize parameters and noise process."""
        self.size = size
        self.mu = mu * np.ones(size)
        self.theta = theta
        self.sigma = sigma
        self.seed = random.seed(seed)
        self.reset()

    def reset(self):
        """Reset the internal state (= noise) to mean (mu)."""
        self.state = copy.copy(self.mu)

    def sample(self):
        """Update internal state and return it as a noise sample."""
        x = self.state
        #dx = self.theta * (self.mu - x) + self.sigma * np.array([random.random() for i in range(len(x))])
        dx = self.theta * (self.mu - x) + self.sigma * np.random.standard_normal(self.size)
        self.state = x + dx
        return self.state

class ReplayBuffer:
    """Fixed-size buffer to store experience tuples."""

    def __init__(self, config):
        """Initialize a ReplayBuffer object.
        Params
        ======
            buffer_size (int): maximum size of buffer
            batch_size (int): size of each training batch
        """
        self.action_size = config.env.action_size
        self.memory = deque(maxlen=config.hp.buffer_size)  # internal memory (deque)
        self.batch_size = config.hp.batch_size
        self.experience = namedtuple("Experience", field_names=["state", "action", "reward", "next_state", "done"])
        self.seed = random.seed(config.general.seed)
        self.device = config.general.device
    
    def add(self, state, action, reward, next_state, done):
        """Add a new experience to memory."""
        e = self.experience(state, action, reward, next_state, done)
        self.memory.append(e)
    
    def sample(self):
        """Randomly sample a batch of experiences from memory."""
        experiences = random.sample(self.memory, k=self.batch_size)

        states = torch.from_numpy(np.vstack([e.state for e in experiences if e is not None])).float().to(self.device)
        actions = torch.from_numpy(np.vstack([e.action for e in experiences if e is not None])).float().to(self.device)
        rewards = torch.from_numpy(np.vstack([e.reward for e in experiences if e is not None])).float().to(self.device)
        next_states = torch.from_numpy(np.vstack([e.next_state for e in experiences if e is not None])).float().to(self.device)
        dones = torch.from_numpy(np.vstack([e.done for e in experiences if e is not None]).astype(np.uint8)).float().to(self.device)

        return (states, actions, rewards, next_states, dones)

    def __len__(self):
        """Return the current size of internal memory."""
        return len(self.memory)