TaskAllocation/allocation_Almighty.py at main · yk7333/TaskAllocation · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
import numpy as np
from manager import Manager,ManagerReplayBuffer
from preAssign import PreAssign,PreAssignReplayBuffer

class Allocation:
    def __init__(self,args):
        self.epoch = 0
        self.manager_buffer_size = args.manager_buffer_size
        self.manager_min_size = args.manager_min_size
        self.manager_batch_size = args.manager_batch_size
        self.preassign_buffer_size = args.preassign_buffer_size
        self.preassign_min_size = args.preassign_min_size
        self.preassign_batch_size = args.preassign_batch_size
        self.hidden_dim = args.hidden_dim
        self.max_epoch = args.max_epoch
        self.gamma = args.gamma
        self.alpha = args.alpha
        self.actor_lr = args.actor_lr
        self.critic_lr = args.critic_lr
        self.tau = args.tau
        self.sigma = args.worker_sigma
        self.sigma_decay = args.sigma_decay_rate
        self.device = args.device
        self.model_dir = args.model_dir
        self.agent_num = args.agent_num
        self.agent_dim = args.agent_dim
        self.task_num = args.task_num
        self.task_dim = args.task_dim
        self.avail_agent = [np.zeros(self.agent_num,dtype=np.int8) for _ in range(self.task_num)]
        self.agent_type = args.agent_type
        self.preassign_buffer = PreAssignReplayBuffer(self.preassign_buffer_size)
        self.preassign = PreAssign(self.task_dim,self.agent_dim,self.hidden_dim,self.actor_lr,self.critic_lr,self.alpha,self.tau,self.gamma,self.device,self.model_dir)
        self.manager_buffer = ManagerReplayBuffer(self.manager_buffer_size)
        self.manager = Manager(self.task_dim,self.agent_dim,self.hidden_dim,self.actor_lr,self.critic_lr,self.alpha,self.tau,self.gamma,self.device,self.agent_type,self.model_dir)


    def chose_agents(self, task_id, task,evaluate,render=True):
        avail_agents = self.avail_agent[task_id]
        state = task.copy()
        done = 0
        returns = 0
        action_list = []
        while not done:
            action = self.manager.select(state,avail_action=avail_agents,evaluate=evaluate)
            next_state,reward,done,avail_agents = self.manager_step(action,task_id)
            transition_dict = {'states': state.copy(),'agent_type':self.agent_type.copy(),'actions': action,'next_states': next_state.copy(),'rewards': reward,'dones': done,'avail_agents':avail_agents}
            self.manager.update(transition_dict)
            state = next_state
            returns += reward
            action_list.append(action)
        if returns<=0:
            action_list=[]
            returns = 0
        if render:
            if len(self.manager.actor_loss_list)>0:
                print("manager{} actor loss: {}".format(task_id,self.manager.actor_loss_list[-1])," manager{} critic loss: {}".format(task_id,self.manager.critic_loss_list[-1]))
                print("manager{} accepts this task!".format(task_id)," allocation",action_list," return:",returns)
        return action_list,returns


    def select(self,state,evaluate=False,render=False):
        self.task = state[:self.task_num*self.task_dim].reshape(self.task_num,self.task_dim)
        self.task_copy = self.task.copy()
        avail_task = (self.task.sum(axis=1)!=0)
        allocation = [[] for _ in range(self.task_num)]
        if sum(avail_task)==0:
            return allocation
        self.agent_type = state[self.task_num*self.task_dim:self.task_num*self.task_dim+self.agent_num*self.agent_dim].reshape(self.agent_num,self.agent_dim)
        avail_agents = state[self.task_num*self.task_dim+self.agent_num*self.agent_dim:]
        self.avail_agent = np.array(self.preassign.select(self.task.copy(),self.agent_type,avail_agents,avail_task,evaluate).T)
        pre_assign = self.avail_agent.copy()
        total_returns = 0
        for task_id,task in enumerate(self.task):
            if avail_task[task_id]==1:
                action_list,returns = self.chose_agents(task_id,task,evaluate,render)
                allocation[task_id].extend(action_list)
                total_returns += returns

        self.preassign_buffer.add(self.task_copy.copy(),pre_assign.T,total_returns,self.agent_type.copy(),avail_agents,avail_task)
        if self.preassign_buffer.size() > self.preassign_min_size:
            pre_states, pre_actions, pre_rewards, pre_agent_types,pre_avail_agents,pre_avail_tasks = self.preassign_buffer.sample(self.preassign_batch_size)
            transition_dict = {'states': pre_states,'actions': pre_actions,'rewards': pre_rewards,"agent_types":pre_agent_types,"avail_agents":pre_avail_agents,"avail_tasks":pre_avail_tasks}
            self.preassign.update(transition_dict)
        return allocation


    def manager_step(self,action,task_id):
        true_action = self.agent_type[action]
        self.task[task_id][1:] = self.task[task_id][1:] - true_action[1:]
        self.task[task_id][1:] = np.where(self.task[task_id][1:]<=0,0,self.task[task_id][1:])
        self.avail_agent[task_id][action] = 0
        done = 0
        if np.all(self.task[task_id][1:]==0) or (sum(self.avail_agent[task_id])==0):
            done = 1
        reward = self.task[task_id][0]-true_action[0] if np.all(self.task[task_id][1:]==0) else -true_action[0]
        return self.task[task_id].copy(), reward, done, self.avail_agent[task_id].copy()

    def save(self):
        self.manager.save()
        self.preassign.save()

    def load(self):
        self.manager.load()
        self.preassign.load()

class AllocationWoPre:
    def __init__(self,args):
        self.epoch = 0
        self.manager_buffer_size = args.manager_buffer_size
        self.manager_min_size = args.manager_min_size
        self.manager_batch_size = args.manager_batch_size
        self.worker_buffer_size = args.worker_buffer_size
        self.worker_min_size = args.worker_min_size
        self.worker_batch_size = args.worker_batch_size
        self.preassign_buffer_size = args.preassign_buffer_size
        self.preassign_min_size = args.preassign_min_size
        self.preassign_batch_size = args.preassign_batch_size
        self.hidden_dim = args.hidden_dim
        self.max_epoch = args.max_epoch
        self.gamma = args.gamma
        self.alpha = args.alpha
        self.actor_lr = args.actor_lr
        self.critic_lr = args.critic_lr
        self.tau = args.tau
        self.sigma = args.worker_sigma
        self.sigma_decay = args.sigma_decay_rate
        self.device = args.device
        self.evaluate = args.evaluate
        self.model_dir = args.model_dir
        self.agent_num = args.agent_num
        self.agent_dim = args.agent_dim
        self.random = args.random
        self.task_num = args.task_num
        self.task_dim = args.task_dim
        self.avail_agent = np.zeros(self.agent_num,dtype=np.int8)
        self.worker_list = []
        self.agent_type = args.agent_type
        self.preassign_buffer = PreAssignReplayBuffer(self.preassign_buffer_size)
        self.preassign = PreAssign(self.task_dim,self.agent_dim,self.hidden_dim,self.actor_lr,self.critic_lr,self.alpha,self.tau,self.gamma,self.device,self.model_dir)
        self.manager_buffer = ManagerReplayBuffer(self.manager_buffer_size)
        self.manager = Manager(self.task_dim,self.agent_dim,self.hidden_dim,self.actor_lr,self.critic_lr,self.alpha,self.tau,self.gamma,self.device,self.agent_type,self.model_dir)


    def chose_agents(self, task_id, task):
        avail_agents = self.avail_agent
        state = task.copy()
        done = 0
        returns = 0
        action_list = []
        while not done:
            action = self.manager.select(state,avail_action=avail_agents)
            next_state,reward,done,avail_agents = self.manager_step(action,task_id)
            transition_dict = {'states': state.copy(),'agent_type':self.agent_type.copy(),'actions': action,'next_states': next_state.copy(),'rewards': reward,'dones': done,'avail_agents':avail_agents}
            self.manager.update(transition_dict)
            state = next_state
            returns += reward
            action_list.append(action)
        if returns<=0:
            action_list=[]
            returns = 0
        else:
            self.avail_agent = avail_agents
        return action_list


    def select(self,state):
        self.task = state[:self.task_num*self.task_dim].reshape(self.task_num,self.task_dim)
        self.task_copy = self.task.copy()
        self.agent_type = state[self.task_num*self.task_dim:self.task_num*self.task_dim+self.agent_num*self.agent_dim].reshape(self.agent_num,self.agent_dim)
        avail_agents = state[self.task_num*self.task_dim+self.agent_num*self.agent_dim:]
        avail_idx = np.where(avail_agents==1)[0]
        self.avail_agent = np.zeros(self.agent_num,dtype=np.int8)
        self.avail_agent[avail_idx] = 1
        allocation = [[] for _ in range(self.task_num)]
        if self.random:
            shuffled_indices = np.random.permutation(self.task.shape[0])
            shuffle_task = self.task[shuffled_indices]
            for t_id,task in enumerate(shuffle_task):
                task_id = shuffled_indices[t_id]
                if sum(task)==0:
                    continue
                allocation[task_id].extend(self.chose_agents(task_id,task))
        else:
            for task_id,task in enumerate(self.task):
                if sum(task)==0:
                    continue
                allocation[task_id].extend(self.chose_agents(task_id,task))
        return allocation


    def manager_step(self,action,task_id):
        true_action = self.agent_type[action]
        self.task[task_id][1:] = self.task[task_id][1:] - true_action[1:]
        self.task[task_id] = np.where(self.task[task_id]<=0,0,self.task[task_id])
        self.avail_agent[action] = 0
        done = 0
        if np.all(self.task[task_id][1:]==0) or (sum(self.avail_agent)==0):
            done = 1
        reward = self.task[task_id][0]-true_action[0] if np.all(self.task[task_id][1:]==0) else -true_action[0]
        return self.task[task_id].copy(), reward, done, self.avail_agent.copy()

    def save(self):
        self.manager.save()
        self.preassign.save()