from typing import Tuple, NamedTuple from collections import namedtuple, deque import numpy as np import random import torch import torch.nn as nn import torch.nn.functional as F from stable_baselines3.common.utils import polyak_update class Experience(NamedTuple): observation: np.ndarray next_observation: np.ndarray action: int reward: float done : bool priority: float = 1 info : dict = {} class BaseBuffer: def __init__(self, size: int): self.size = size self.experience = deque(maxlen=size) def __len__(self): return len(self.experience) def add(self, experience): self.experience.append(experience) def sample(self, k): sample = random.choices(self.experience, k=k) return sample class PERBuffer(BaseBuffer): def __init__(self, size, alpha=0.2): super(PERBuffer, self).__init__(size) self.alpha = alpha def sample(self, k): pr = [abs(e.priority)**self.alpha for e in self.experience] pr = np.array(pr) / sum(pr) idxs = random.choices(range(len(self)), weights=pr, k=k) pass class BaseDQN(nn.Module): def __init__(self): super(BaseDQN, self).__init__() self.net = nn.Sequential( nn.Linear(3*5*5, 64), nn.ReLU(), nn.Linear(64, 64), nn.ReLU(), nn.Linear(64, 9) ) def act(self, x): with torch.no_grad(): action = self.net(x.view(x.shape[0], -1)).argmax(-1).item() return action def forward(self, x): return self.net(x.view(x.shape[0], -1)) def random_action(self): return random.randrange(0, 5) class BaseQlearner: def __init__(self, q_net, target_q_net, env, buffer, target_update, warmup, eps_end, gamma=0.99, train_every_n_steps=4, n_grad_steps=1, exploration_fraction=0.2, batch_size=64, lr=1e-4, reg_weight=0.0): self.q_net = q_net self.target_q_net = target_q_net self.target_q_net.load_state_dict(self.q_net.state_dict()) self.target_q_net.eval() self.env = env self.buffer = buffer self.target_update = target_update self.warmup = warmup self.eps = 1. self.eps_end = eps_end self.exploration_fraction = exploration_fraction self.batch_size = batch_size self.gamma = gamma self.train_every_n_steps = train_every_n_steps self.n_grad_steps = n_grad_steps self.lr = lr self.reg_weight = reg_weight self.optimizer = torch.optim.Adam(self.q_net.parameters(), lr=self.lr) self.device = 'cpu' self.running_reward = deque(maxlen=10) self.running_loss = deque(maxlen=10) def to(self, device): self.device = device return self def anneal_eps(self, step, n_steps): fraction = min(float(step) / int(self.exploration_fraction*n_steps), 1.0) eps = 1 + fraction * (self.eps_end - 1) return eps def learn(self, n_steps): step, eps = 0, 1 while step < n_steps: obs, done = self.env.reset(), False total_reward = 0 while not done: action = self.q_net.act(torch.from_numpy(obs).unsqueeze(0).float()) \ if np.random.rand() > eps else env.action_space.sample() next_obs, reward, done, info = env.step(action) experience = Experience(observation=obs, next_observation=next_obs, action=action, reward=reward, done=done) # do we really need to copy? self.buffer.add(experience) # end of step routine obs = next_obs step += 1 total_reward += reward eps = self.anneal_eps(step, n_steps) if step % self.train_every_n_steps == 0: self.train() if step % self.target_update == 0: self.target_q_net.load_state_dict(self.q_net.state_dict()) self.running_reward.append(total_reward) if step % 10 == 0: print(f'Step: {step} ({(step/n_steps)*100:.2f}%)\tRunning reward: {sum(list(self.running_reward))/len(self.running_reward)}\t' f' eps: {eps:.4f}\tRunning loss: {sum(list(self.running_loss))/len(self.running_loss)}') def train(self): for _ in range(self.n_grad_steps): experience = self.buffer.sample(self.batch_size) obs = torch.stack([torch.from_numpy(e.observation) for e in experience], 0).float() next_obs = torch.stack([torch.from_numpy(e.next_observation) for e in experience], 0).float() actions = torch.tensor([e.action for e in experience]).long() rewards = torch.tensor([e.reward for e in experience]).float() dones = torch.tensor([e.done for e in experience]).float() next_q_values = self.target_q_net(next_obs).detach().max(-1)[0] target_q_values = rewards + (1. - dones) * self.gamma * next_q_values q_values = self.q_net(obs).gather(-1, actions.unsqueeze(0)) delta = q_values - target_q_values loss = torch.mean(self.reg_weight * q_values + torch.pow(delta, 2)) self.running_loss.append(loss.item()) # Optimize the model self.optimizer.zero_grad() loss.backward() torch.nn.utils.clip_grad_norm_(self.q_net.parameters(), 10) self.optimizer.step() if __name__ == '__main__': from environments.factory.simple_factory import SimpleFactory, DirtProperties, MovementProperties from algorithms.reg_dqn import RegDQN dirt_props = DirtProperties(clean_amount=3, gain_amount=0.2, max_global_amount=30, max_local_amount=5, spawn_frequency=1, max_spawn_ratio=0.05) move_props = MovementProperties(allow_diagonal_movement=True, allow_square_movement=True, allow_no_op=False) env = SimpleFactory(dirt_properties=dirt_props, movement_properties=move_props, n_agents=1, pomdp_radius=2, max_steps=400, omit_agent_slice_in_obs=False) #print(env.action_space) from stable_baselines3.dqn import DQN #dqn = RegDQN('MlpPolicy', env, verbose=True, buffer_size = 50000, learning_starts = 25000, batch_size = 64, target_update_interval = 5000, exploration_fraction = 0.25, exploration_final_eps = 0.025) #print(dqn.policy) #dqn.learn(100000) print(env.observation_space, env.action_space) dqn, target_dqn = BaseDQN(), BaseDQN() learner = BaseQlearner(dqn, target_dqn, env, BaseBuffer(50000), target_update=5000, warmup=25000, lr=1e-4, gamma=0.99, train_every_n_steps=4, eps_end=0.05, reg_weight=0.1, exploration_fraction=0.25, batch_size=64) learner.learn(100000)