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

# BUFFER_SIZE = int(1e5)  # replay buffer size
# BATCH_SIZE = 256        # minibatch size
# TAU = 1e-3              # for soft update of target parameters
LR_ACTOR = 1e-4         # learning rate of the actor 
LR_CRITIC = 3e-4        # learning rate of the critic
WEIGHT_DECAY = 0        # L2 weight decay

NOISE_REDUCTION_RATE = 0.99
EPISODES_BEFORE_TRAINING = 500
NOISE_START=1.0
NOISE_END=0.1

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

class Agent():
    """Interacts with and learns from the environment."""
    
    def __init__(self, state_size, action_size, num_agents, random_seed):
        """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.state_size = state_size
        self.action_size = action_size
        self.seed = random.seed(random_seed)

        # Actor Network (w/ Target Network)
        self.actor_local = Actor(state_size, action_size, random_seed).to(device)
        self.actor_target = Actor(state_size, action_size, random_seed).to(device)
        self.actor_optimizer = optim.Adam(self.actor_local.parameters(), lr=LR_ACTOR)

        # Critic Network (w/ Target Network)
        self.critic_local = Critic(state_size*num_agents, action_size*num_agents, random_seed).to(device)
        self.critic_target = Critic(state_size*num_agents, action_size*num_agents, random_seed).to(device)
        self.critic_optimizer = optim.Adam(self.critic_local.parameters(), lr=LR_CRITIC, weight_decay=WEIGHT_DECAY)

        self.soft_update(self.critic_local, self.critic_target, 1)
        self.soft_update(self.actor_local, self.actor_target, 1)    
        
        # Noise process
        self.noise = OUNoise(action_size, random_seed)
        self.noise_reduction_ratio = NOISE_START

        self.step_count = 0

    def act(self, state, i_episode, add_noise=True):
        """Returns actions for given state as per current policy."""
        state = torch.from_numpy(state).float().to(device)
        self.actor_local.eval()
        with torch.no_grad():
            action = self.actor_local(state).cpu().data.numpy()
        self.actor_local.train()
        
        if add_noise:
            if i_episode > EPISODES_BEFORE_TRAINING and self.noise_reduction_ratio > NOISE_END:
                self.noise_reduction_ratio = NOISE_REDUCTION_RATE**(i_episode-EPISODES_BEFORE_TRAINING)
#             noise_reduction_ratio = 1
            action += self.noise_reduction_ratio * self.add_noise2()
#             action += noise_reduction_ratio * self.noise.sample()
        return np.clip(action, -1, 1)

    def add_noise2(self):
#         noise = 0.5*np.random.randn(1,self.action_size) #sigma of 0.5 as sigma of 1 will have alot of actions just clipped
        noise = 0.5*np.random.standard_normal(self.action_size)
        return noise
    
    def reset(self):
        self.noise.reset()

    def learn(self, experiences, gamma):
        """Update policy and value parameters using given batch of experience tuples.
        Q_targets = r + γ * critic_target(next_state, actor_target(next_state))
        where:
            actor_target(state) -> action
            critic_target(state, action) -> Q-value
        Params
        ======
            experiences (Tuple[torch.Tensor]): tuple of (s, a, r, s', done) tuples 
            gamma (float): discount factor
        """
        full_states, actions, actor_local_actions, actor_target_actions, agent_state, agent_action, agent_reward, agent_done, next_states, next_full_states = experiences

        # ---------------------------- update critic ---------------------------- #
        # Get predicted next-state actions and Q values from target models
#         actions_next = self.actor_target(next_states)
        Q_targets_next = self.critic_target(next_full_states, actor_target_actions)
        # Compute Q targets for current states (y_i)
        Q_targets = agent_reward + (gamma * Q_targets_next * (1 - agent_done))
        # Compute critic loss
        Q_expected = self.critic_local(full_states, actions)
        critic_loss = F.mse_loss(Q_expected, Q_targets)
        # Minimize the loss
        self.critic_optimizer.zero_grad()
        critic_loss.backward()
#         torch.nn.utils.clip_grad_norm_(self.critic_local.parameters(), 1)
        self.critic_optimizer.step()

        # ---------------------------- update actor ---------------------------- #
        # Compute actor loss
#         actions_pred = self.actor_local(agent_state)
        actor_loss = -self.critic_local(full_states, actor_local_actions).mean()
        # Minimize the loss
        self.actor_optimizer.zero_grad()
        actor_loss.backward()
        self.actor_optimizer.step()

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

    def hard_copy_weights(self, target, source):
        """ copy weights from source to target network (part of initialization)"""
        for target_param, param in zip(target.parameters(), source.parameters()):
            target_param.data.copy_(param.data)
            
    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=0.1):
        """Initialize parameters and noise process."""
        self.mu = mu * np.ones(size)
        self.theta = theta
        self.sigma = sigma
        self.seed = random.seed(seed)
        self.reset()
        self.size = size

    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