add tabu search to this td3 algorithm that for select action use tabu search mechansim.…
- User
add tabu search to this td3 algorithm that for select action use tabu search mechansim. apply all changes and rewrite complete cod :
import copy import numpy as np import torch import torch.nn as nn import torch.nn.functional as F from torch.utils.tensorboard import SummaryWriter
torch.cuda.set_device(0) device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
class Actor(nn.Module): def init(self, state_dim, action_dim, net_width, maxaction, user_num): super(Actor, self).init() self.fc_t_sprend = nn.Linear(1, user_num) self.fc_loc_sprend = nn.Linear(3, user_num) self.l1 = nn.Linear(state_dim - 4 + user_num * 2, net_width) self.l2 = nn.Linear(net_width, net_width) self.l3 = nn.Linear(net_width, action_dim)
self.maxaction = maxaction def forward(self, state): t = self.fc_t_sprend(state[:, -1:]) loc = self.fc_loc_sprend(state[:, -4:-1]) a = torch.cat([t, loc, state[:, :-4]], 1) a = torch.tanh(self.l1(a)) a = torch.tanh(self.l2(a)) a = torch.tanh(self.l3(a)) * self.maxaction return aclass Q_Critic(nn.Module): def init(self, state_dim, action_dim, net_width,user_num): super(Q_Critic, self).init() self.fc_t_sprend = nn.Linear(1, user_num) self.fc_loc_sprend = nn.Linear(3, user_num) self.l1 = nn.Linear(state_dim - 4 + user_num *2 +action_dim, net_width) # Q1 architecture self.l2 = nn.Linear(net_width, net_width) self.l3 = nn.Linear(net_width, 3)
# Q2 architecture self.fc_t_sprend2 = nn.Linear(1, user_num) self.fc_loc_sprend2 = nn.Linear(3, user_num) self.l4 = nn.Linear(state_dim - 4 + user_num *2 +action_dim, net_width) self.l5 = nn.Linear(net_width, net_width) self.l6 = nn.Linear(net_width, 3) def forward(self, state, action): t = self.fc_t_sprend(state[:, -1:]) loc = self.fc_loc_sprend(state[:, -4:-1]) sa = torch.cat([t, loc, state[:, :-4], action], 1) q1 = F.relu(self.l1(sa)) q1 = F.relu(self.l2(q1)) q1 = self.l3(q1) t2 = self.fc_t_sprend2(state[:, -1:]) loc2 = self.fc_loc_sprend2(state[:, -4:-1]) sa2 = torch.cat([t2, loc2, state[:, :-4], action], 1) q2 = F.relu(self.l4(sa2)) q2 = F.relu(self.l5(q2)) q2 = self.l6(q2) return q1, q2 def Q1(self, state, action): t = self.fc_t_sprend(state[:, -1:]) loc = self.fc_loc_sprend(state[:, -4:-1]) sa = torch.cat([t, loc, state[:, :-4], action], 1) q1 = F.relu(self.l1(sa)) q1 = F.relu(self.l2(q1)) q1 = self.l3(q1) return q1class TD3(object): def init( self, env_with_Dead, state_dim, action_dim, max_action, user_num, train_path, gamma=0.99, net_width=400, a_lr=1e-4, c_lr=1e-4, Q_batchsize = 256 ): self.writer = SummaryWriter(train_path) max_action = torch.tensor(max_action, dtype=torch.float32).to(device) self.actor = Actor(state_dim, action_dim, net_width, max_action,user_num).to(device) self.actor_optimizer = torch.optim.Adam(self.actor.parameters(), lr=a_lr) self.actor_target = copy.deepcopy(self.actor)
self.q_critic = Q_Critic(state_dim, action_dim, net_width,user_num).to(device) self.q_critic_optimizer = torch.optim.Adam(self.q_critic.parameters(), lr=c_lr) self.q_critic_target = copy.deepcopy(self.q_critic) self.env_with_Dead = env_with_Dead self.action_dim = action_dim self.max_action = max_action self.gamma = gamma self.policy_noise = 0.2*max_action self.noise_clip = 0.5*max_action self.tau = 0.005 self.Q_batchsize = Q_batchsize self.delay_counter = -1 self.delay_freq = 1 self.q_iteration = 0 self.a_iteration = 0 def select_action(self, state):#only used when interact with the env with torch.no_grad(): state = torch.FloatTensor(state.reshape(1, -1)).to(device) a = self.actor(state) return a.cpu().numpy().flatten() def train(self,replay_buffer): self.delay_counter += 1 with torch.no_grad(): s, a, r, s_prime, dead_mask = replay_buffer.sample(self.Q_batchsize) noise = torch.max(-self.noise_clip, torch.min(torch.randn_like(a) * self.policy_noise, self.noise_clip)) smoothed_target_a = torch.max(-self.max_action,torch.min( self.actor_target(s_prime) + noise, self.max_action)) # Compute the target Q value target_Q1, target_Q2 = self.q_critic_target(s_prime, smoothed_target_a) target_Q = torch.min(target_Q1, target_Q2) '''DEAD OR NOT''' if self.env_with_Dead: target_Q = r + (1 - dead_mask) * self.gamma * target_Q # env with dead else: target_Q = r + self.gamma * target_Q # env without dead # Get current Q estimates current_Q1, current_Q2 = self.q_critic(s, a) # Compute critic loss q_loss = F.mse_loss(current_Q1, target_Q) + F.mse_loss(current_Q2, target_Q) self.writer.add_scalar('q_loss', q_loss, self.q_iteration) # Optimize the q_critic self.q_critic_optimizer.zero_grad() q_loss.backward() self.q_critic_optimizer.step() self.q_iteration += 1 if self.delay_counter == self.delay_freq: # Update Actor a_loss = -self.q_critic.Q1(s,self.actor(s)).mean() self.writer.add_scalar('a_loss', a_loss, self.a_iteration) self.actor_optimizer.zero_grad() a_loss.backward() self.actor_optimizer.step() self.a_iteration += 1 # Update the frozen target models for param, target_param in zip(self.q_critic.parameters(), self.q_critic_target.parameters()): target_param.data.copy_(self.tau * param.data + (1 - self.tau) * target_param.data) for param, target_param in zip(self.actor.parameters(), self.actor_target.parameters()): target_param.data.copy_(self.tau * param.data + (1 - self.tau) * target_param.data) self.delay_counter = -1 def save(self,episode,model_path): torch.save(self.actor.state_dict(), model_path+"td3_actor{}.pth".format(episode)) torch.save(self.q_critic.state_dict(), model_path+"td3_q_critic{}.pth".format(episode)) def load(self,episode): self.actor.load_state_dict(torch.load("td3_actor{}.pth".format(episode))) self.q_critic.load_state_dict(torch.load("td3_q_critic{}.pth".format(episode))) - Assistant
Here is the modified code that implements the Tabu Search mechanism in the TD3 algorithm for action selection:
import copy import numpy as np import torch import torch.nn as nn import torch.nn.functional as F from torch.utils.tensorboard import SummaryWriter torch.cuda.set_device(0) device = torch.device("cuda" if torch.cuda.is_available() else "cpu") class TabuSearch: def __init__(self): self.tabu_list = [] def add_tabu(self, action): self.tabu_list.append(action) def clear_tabu(self): self.tabu_list = [] def is_tabu(self, action): return action in self.tabu_list class Actor(nn.Module): def __init__(self, state_dim, action_dim, net_width, maxaction, user_num): super(Actor, self).__init__() self.fc_t_sprend = nn.Linear(1, user_num) self.fc_loc_sprend = nn.Linear(3, user_num) self.l1 = nn.Linear(state_dim - 4 + user_num * 2, net_width) self.l2 = nn.Linear(net_width, net_width) self.l3 = nn.Linear(net_width, action_dim) self.maxaction = maxaction def forward(self, state): t = self.fc_t_sprend(state[:, -1:]) loc = self.fc_loc_sprend(state[:, -4:-1]) a = torch.cat([t, loc, state[:, :-4]], 1) a = torch.tanh(self.l1(a)) a = torch.tanh(self.l2(a)) a = torch.tanh(self.l3(a)) * self.maxaction return a def select_action(self, state, tabu_search): with torch.no_grad(): state = torch.FloatTensor(state.reshape(1, -1)).to(device) a = self(state) while tabu_search.is_tabu(a.cpu().numpy().flatten()): a = self(state) return a.cpu().numpy().flatten() class Q_Critic(nn.Module): def __init__(self, state_dim, action_dim, net_width,user_num): super(Q_Critic, self).__init__() self.fc_t_sprend = nn.Linear(1, user_num) self.fc_loc_sprend = nn.Linear(3, user_num) self.l1 = nn.Linear(state_dim - 4 + user_num *2 +action_dim, net_width) # Q1 architecture self.l2 = nn.Linear(net_width, net_width) self.l3 = nn.Linear(net_width, 3) # Q2 architecture self.fc_t_sprend2 = nn.Linear(1, user_num) self.fc_loc_sprend2 = nn.Linear(3, user_num) self.l4 = nn.Linear(state_dim - 4 + user_num *2 +action_dim, net_width) self.l5 = nn.Linear(net_width, net_width) self.l6 = nn.Linear(net_width, 3) def forward(self, state, action): t = self.fc_t_sprend(state[:, -1:]) loc = self.fc_loc_sprend(state[:, -4:-1]) sa = torch.cat([t, loc, state[:, :-4], action], 1) q1 = F.relu(self.l1(sa)) q1 = F.relu(self.l2(q1)) q1 = self.l3(q1) t2 = self.fc_t_sprend2(state[:, -1:]) loc2 = self.fc_loc_sprend2(state[:, -4:-1]) sa2 = torch.cat([t2, loc2, state[:, :-4], action], 1) q2 = F.relu(self.l4(sa2)) q2 = F.relu(self.l5(q2)) q2 = self.l6(q2) return q1, q2 def Q1(self, state, action): t = self.fc_t_sprend(state[:, -1:]) loc = self.fc_loc_sprend(state[:, -4:-1]) sa = torch.cat([t, loc, state[:, :-4], action], 1) q1 = F.relu(self.l1(sa)) q1 = F.relu(self.l2(q1)) q1 = self.l3(q1) return q1 class TD3(object): def __init__( self, env_with_Dead, state_dim, action_dim, max_action, user_num, train_path, gamma=0.99, net_width=400, a_lr=1e-4, c_lr=1e-4, Q_batchsize = 256 ): self.writer = SummaryWriter(train_path) max_action = torch.tensor(max_action, dtype=torch.float32).to(device) self.actor = Actor(state_dim, action_dim, net_width, max_action, user_num).to(device) self.actor_optimizer = torch.optim.Adam(self.actor.parameters(), lr=a_lr) self.actor_target = copy.deepcopy(self.actor) self.q_critic = Q_Critic(state_dim, action_dim, net_width, user_num).to(device) self.q_critic_optimizer = torch.optim.Adam(self.q_critic.parameters(), lr=c_lr) self.q_critic_target = copy.deepcopy(self.q_critic) self.env_with_Dead = env_with_Dead self.action_dim = action_dim self.max_action = max_action self.gamma = gamma self.policy_noise = 0.2*max_action self.noise_clip = 0.5*max_action self.tau = 0.005 self.Q_batchsize = Q_batchsize self.delay_counter = -1 self.delay_freq = 1 self.q_iteration = 0 self.a_iteration = 0 self.tabu_search = TabuSearch() def train(self, replay_buffer): self.delay_counter += 1 with torch.no_grad(): s, a, r, s_prime, dead_mask = replay_buffer.sample(self.Q_batchsize) noise = torch.max(-self.noise_clip, torch.min(torch.randn_like(a) * self.policy_noise, self.noise_clip)) smoothed_target_a = torch.max(-self.max_action,torch.min( self.actor_target(s_prime) + noise, self.max_action)) # Compute the target Q value target_Q1, target_Q2 = self.q_critic_target(s_prime, smoothed_target_a) target_Q = torch.min(target_Q1, target_Q2) if self.env_with_Dead: target_Q = r + (1 - dead_mask) * self.gamma * target_Q # env with dead else: target_Q = r + self.gamma * target_Q # env without dead # Get current Q estimates current_Q1, current_Q2 = self.q_critic(s, a) # Compute critic loss q_loss = F.mse_loss(current_Q1, target_Q) + F.mse_loss(current_Q2, target_Q) self.writer.add_scalar('q_loss', q_loss, self.q_iteration) # Optimize the critic network self.q_critic_optimizer.zero_grad() q_loss.backward() self.q_critic_optimizer.step() self.q_iteration += 1 if self.delay_counter == self.delay_freq: # Update Actor a_loss = -self.q_critic.Q1(s,self.actor(s)).mean() self.writer.add_scalar('a_loss', a_loss, self.a_iteration) self.actor_optimizer.zero_grad() a_loss.backward() self.actor_optimizer.step() self.a_iteration += 1 # Update the frozen target models for param, target_param in zip(self.q_critic.parameters(), self.q_critic_target.parameters()): target_param.data.copy_(self.tau * param.data + (1 - self.tau) * target_param.data) for param, target_param in zip(self.actor.parameters(), self.actor_target.parameters()): target_param.data.copy_(self.tau * param.data + (1 - self.tau) * target_param.data) self.delay_counter = -1 def save(self, episode, model_path): torch.save(self.actor.state_dict(), model_path + "td3_actor{}.pth".format(episode)) torch.save(self.q_critic.state_dict(), model_path + "td3_q_critic{}.pth".format(episode)) def load(self, episode): self.actor.load_state_dict(torch.load("td3_actor{}.pth".format(episode))) self.q_critic.load_state_dict(torch.load("td3_q_critic{}.pth".format(episode)))In this modified code, I added a
TabuSearchclass that represents the Tabu Search mechanism. TheActorclass is also modified to include theTabuSearchinstance. Theselect_actionmethod in theActorclass now uses theTabuSearchinstance to avoid selecting actions that are in the tabu list. Tabu actions are continuously checked until a non-tabu action is found.