論文解讀?論文筆記 Hierarchical Reinforcement Learning for Scarce Medical Resource Allocation_UQI-LIUWJ的博客-CSDN博客

代碼部分

KYHKL-Q/Hierarchical-RL (github.com)?

1 RL_train.py

訓(xùn)練強(qiáng)化學(xué)習(xí)模塊的部分

由于論文中提及了口罩和床位兩個(gè)指標(biāo)，兩個(gè)的模型在很多方面都是類似的，所以為了方便起見，我只設(shè)計(jì)口罩部分

1.1 導(dǎo)入庫

import numpy as np import json import torch import random import copy import os from torch import nn,optim from simulator import simulator from reward_fun import reward_fun

1.2 RL_train 類

1.2.1 __init__

def __init__(self,action_lr=0.001, #action ，也就是論文中顯著性排序DQN的學(xué)習(xí)率Actor_lr=0.001, #actor（生成滿足因子）的學(xué)習(xí)率Critic_lr=0.002, #critic的學(xué)習(xí)率decay_rate=0.5, #未來獎(jiǎng)勵(lì)的折現(xiàn)率 batch_size=8, #batch_sizepool_volume=128, #經(jīng)驗(yàn)回放 buffer的大小action_explore_factor=0.5, #action 探索的比例 （ε-greedy）Actor_explore_factor=0.05, #actor探索的比例 （ε-greedy）soft_replace_rate=0.1, #軟更新，eval的參數(shù)對(duì)target net的影響#DDPG問題中，每次target net向eval net靠近的幅度train_steps=150, #總共要訓(xùn)練的輪數(shù)（類似于epoch-number）interval=48, #每隔多久更新一次八狀態(tài)mask_total=1000000, #總的口罩?jǐn)?shù)量mask_quality=0.9, #醫(yī)用口罩有效過濾病毒的比例mask_lasting=48, #口罩有效期city='city_sample' #研究區(qū)域的名稱):print('Initializing...')self.action_lr=action_lrself.Actor_lr=Actor_lrself.Critic_lr=Critic_lrself.decay_rate=decay_rate self.batch_size=batch_sizeself.pool_volume=pool_volumeself.action_explore_factor=action_explore_factorself.Actor_explore_factor=Actor_explore_factorself.soft_replace_rate=soft_replace_rate self.train_steps=train_stepsself.interval=intervalself.mask_total=mask_totalself.mask_quality=mask_qualityself.mask_lasting=mask_lastingself.city=city#Initialize state and replay bufferwith open(os.path.join('../data',self.city,'start.json'),'r') as f:self.start = np.array(json.load(f))#由于提供的代碼里面沒有這個(gè)文件，我猜測(cè)是初始情況下各區(qū)域各狀態(tài)的人數(shù)##是一個(gè)region_num*8的二維數(shù)組from Net.Anet import Anet#actor network————>計(jì)算滿足因子from Net.Qnet import Qnet#action network————>計(jì)算選擇哪個(gè)顯著性排序from Net.RNN import RNN#用可以求得的三個(gè)狀態(tài)'''It（已經(jīng)感染疾病，同時(shí)被檢測(cè)出疾病的人。）Ih（被送醫(yī)治療的感染者）D（死亡人群）'''#計(jì)算其他三個(gè)環(huán)境可看到的狀態(tài)的值'''E（暴露人群）Iu（已經(jīng)感染疾病，但是沒有檢測(cè)的人。）R（康復(fù)人群）'''from Net.Cnet import Cnet#critic networkself.region_num = 161#Supposed to be modified according to the number of regions in the studied city.#區(qū)劃的數(shù)量self.pool_pointer=0#經(jīng)驗(yàn)回放 relay buffer 下一個(gè)下標(biāo)（下一組記錄放到哪里）self.pool_count=0#目前relay buffer中的記錄個(gè)數(shù)self.pool_state=np.zeros([self.pool_volume,self.region_num,8])#relay buffer中的狀態(tài)數(shù)#buffer_size * region_num * 8self.pool_mask_action=np.zeros(self.pool_volume)#relay buffer 中選擇了哪個(gè)顯著性排序self.pool_mask_perc=np.zeros(self.pool_volume)#relay buffer 中 口罩的滿足因子self.pool_reward=np.zeros(self.pool_volume)#relay buffer 中的單步獎(jiǎng)勵(lì)self.pool_state1=np.zeros([self.pool_volume,self.region_num,8])#relay buffer 中的下一狀態(tài)self.current_state=self.start#目前的狀態(tài)self.next_state=self.current_state#下一狀態(tài)#Initialize the networkstorch.manual_seed(1234)torch.cuda.manual_seed(1234)self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')self.Mask_action_eval=Qnet().to(self.device)self.Mask_action_eval.train()self.Mask_action_target=copy.deepcopy(self.Mask_action_eval)#深拷貝，也就是二者不共享參數(shù)#一個(gè)是eval network 一個(gè)是target networkself.Mask_action_target.eval()#表示target 不參與訓(xùn)練（后同） self.Mask_action_optimizer=optim.Adam(self.Mask_action_eval.parameters(),lr=self.action_lr)#action 指的是論文模型示意圖中的DQN，也就是學(xué)習(xí)顯著性排序的部分self.Mask_Actor_eval=Anet().to(self.device)self.Mask_Actor_eval.train()self.Mask_Actor_target=copy.deepcopy(self.Mask_Actor_eval) self.Mask_Actor_target.eval()self.Mask_Actor_optimizer=optim.Adam(self.Mask_Actor_eval.parameters(),lr=self.Actor_lr)#actor 就是論文模型示意圖下面的actorself.Mask_Critic_eval=Cnet(self.region_num).to(self.device)self.Mask_Critic_eval.train()self.Mask_Critic_target=copy.deepcopy(self.Mask_Critic_eval)self.Mask_Critic_eval.eval() self.Mask_Critic_optimizer=optim.Adam(self.Mask_Critic_eval.parameters(),lr=self.Critic_lr)#critic 就是論文模型示意圖下面的critic#Initialize the simulator#各區(qū)域疾病狀態(tài)轉(zhuǎn)換self.simulator=simulator(city=self.city)self.simulator.reset(self.start)#將start中各區(qū)域各狀態(tài)的初始人數(shù)設(shè)置到simulater 的nodes列表中#nodes列表的每一個(gè)元素都是一個(gè)node對(duì)象#node對(duì)象有一個(gè)id，以及8狀態(tài)（S,E,Iu,It,Ia,Ih,R,D）print('Initializing down!')

1.2.2 train

def train(self):print('Start training...')loss_record = list()train_count = 0 #目前訓(xùn)練的輪數(shù)while train_count < self.train_steps:self.current_state=self.start##由于提供的代碼里面沒有這個(gè)文件，我猜測(cè)start是初始情況下各區(qū)域各狀態(tài)的人數(shù)##是一個(gè)region_num*8的二維數(shù)組self.next_state=self.current_state#下一狀態(tài)self.is_end=False#是否結(jié)束episodeend_count = 0#samplingstep_count=0while ((not self.is_end) andstep_count <= 60/(self.interval/48)):mask_action_out = self.Mask_action_eval(torch.FloatTensor(self.current_state[:,[1,2,3,5,6,7]]).to(self.device).unsqueeze(0))'''選取E,Iu,It,Ih,R,D同時(shí)升維度至[1,region_num,6]action 指的是論文模型示意圖中的DQN，也就是學(xué)習(xí)顯著性排序的部分mask 就是指看口罩mask_action_out->[1,7]這里是表示不同的排序原則的“打分”論文中一共出現(xiàn)了3中不同的排序原則，其中可以任意組合，所以有2^3-1=7種（不能都不選）'''#select an action through epsilon-greedymask_action=torch.argmax(mask_action_out).cpu().item()#選擇“打分最高的”顯著性排序原則rand_temp=random.random()if(rand_temp>(1-self.action_explore_factor)):mask_action = int(7 * random.random())#一定比例是隨機(jī)探索,否則就是利用print('mask_action:{}'.format(mask_action))mask_perc = self.Mask_Actor_eval(torch.FloatTensor(self.current_state[:,[1,2,3,5,6,7]]).to(self.device).unsqueeze(0))#選取E,Iu,It,Ih,R,D#同時(shí)升維度至[1,region_num,6]#mask_perc是一個(gè)[1,1]的0~1之間的數(shù)，表示的是論文里提到的滿足因子fmask_perc = mask_perc.squeeze(0).cpu().detach().item()\+ self.Actor_explore_factor * np.random.randn()mask_perc_clip = np.clip(mask_perc, 0.1, 1)print('mask_perc:{}'.format(mask_perc_clip))#我們通過actor算出來的比例，加上一定探索的概率，得到最終的滿足因子f#與此同時(shí)，滿足因子f需要在0.1~1之間#get the next state through simulationself.next_state, _ = self.simulator.simulate(sim_type='Policy_a',interval=self.interval,bed_total=self.bed_total,bed_action=bed_action,bed_satisfy_perc=bed_perc_clip,mask_on=True,mask_total=self.mask_total,mask_quality=self.mask_quality,mask_lasting=self.mask_lasting,mask_action=mask_action,mask_satisfy_perc=mask_perc_clip)#各個(gè)區(qū)域新的狀態(tài)if(self.simulator.full and end_count==0):end_count=1#full表示床位有沒有滿if((not self.simulator.full) and end_count==1):self.is_end=Trueprint('Is_end:{}'.format(self.is_end))#get single step rewardreward=reward_fun(self.current_state,self.next_state)#單步獎(jiǎng)勵(lì)函數(shù)print('Reward:{}'.format(reward))#put the sample into the replay buffer#將數(shù)據(jù)放入經(jīng)驗(yàn)回放中if(self.simulator.full):self.pool_state[self.pool_pointer]=self.current_stateself.pool_bed_action[self.pool_pointer]=bed_actionself.pool_mask_action[self.pool_pointer]=mask_actionself.pool_bed_perc[self.pool_pointer]=bed_percself.pool_mask_perc[self.pool_pointer]=mask_percself.pool_reward[self.pool_pointer]=rewardself.pool_state1[self.pool_pointer]=self.next_stateself.pool_pointer=(self.pool_pointer+1)%self.pool_volumeif (self.pool_count<self.pool_volume):self.pool_count=self.pool_count+1print('Sampling:{} sampeles in pool now'.format(self.pool_count))#將當(dāng)前的狀態(tài)、床位和口罩使用哪種顯著性排序、床位和口罩的滿足因子#單步獎(jiǎng)勵(lì)，下一狀態(tài) 送到relay buffer中#mini batch sampling and updating the parametersif (self.pool_count>=self.batch_size):sample_index=random.sample(range(self.pool_count),self.batch_size)#從relay buffer中隨機(jī)選擇batch_size個(gè)indexsample_state=torch.FloatTensor(self.pool_state[sample_index]).to(self.device)#選擇的batch中的當(dāng)前狀態(tài)sample_bed_action=self.pool_bed_action[sample_index]#選擇的batch中的床位顯著性排序sample_mask_action=self.pool_mask_action[sample_index]#選擇的batch中的口罩顯著性排序sample_reward=self.pool_reward[sample_index]#選擇的batch中的單步獎(jiǎng)勵(lì)sample_bed_perc=torch.FloatTensor(self.pool_bed_perc[sample_index]).to(self.device).unsqueeze(1)#選擇的batch中的床位滿足因子sample_mask_perc=torch.FloatTensor(self.pool_mask_perc[sample_index]).to(self.device).unsqueeze(1)#選擇的batch中的口罩滿足因子sample_state1=torch.FloatTensor(self.pool_state1[sample_index]).to(self.device)##選擇的batch中的下一個(gè)狀態(tài)mask_action_eval=self.Mask_action_eval(sample_state[:,:,[1,2,3,5,6,7]])#顯著性排序的eval_net#sample_state[:,:,[1,2,3,5,6,7]]————>[batch_size,region_num,6]#結(jié)果為[batch_size,7]mask_action_target=self.Mask_action_target(sample_state1[:,:,[1,2,3,5,6,7]])#顯著性排序的target_net#sample_state[:,:,[1,2,3,5,6,7]]————>[batch_size,region_num,6]#結(jié)果為[batch_size,7]mask_action_max = torch.argmax(mask_action_target, dim=-1).cpu()#batch_size 每個(gè)最大的action，也就是bacth中每個(gè)記錄應(yīng)該選擇哪一種顯著性排序mask_perc_eval=self.Mask_Actor_eval(sample_state[:,:,[1,2,3,5,6,7]])#滿足因子的eval_net#sample_state[:,:,[1,2,3,5,6,7]]————>[batch_size,region_num,6]#結(jié)果為[batch_size,1]mask_perc_target=self.Mask_Actor_target(sample_state1[:,:,[1,2,3,5,6,7]])#滿足因子的target_net#sample_state[:,:,[1,2,3,5,6,7]]————>[batch_size,region_num,6]#結(jié)果為[batch_size,1]mask_reward_eval = self.Mask_Critic_eval(sample_state[:,:,[1,2,3,5,6,7]], sample_mask_perc)#critic的eval_net#sample_state[:,:,[1,2,3,5,6,7]]————>[batch_size,region_num,6]#結(jié)果為[batch_size,1]mask_reward_target=self.Mask_Critic_target(sample_state1[:,:,[1,2,3,5,6,7]],mask_perc_target)#critic的eval_net#sample_state[:,:,[1,2,3,5,6,7]]————>[batch_size,region_num,6]#結(jié)果為[batch_size,1]loss = 0for i in range(self.batch_size):y = sample_reward[i]y = y + \self.decay_rate * (mask_action_target[i][mask_action_max[i]] +mask_reward_target[i])#顯著性排序中最大的那個(gè)的值+critic reward時(shí)的值loss = loss + \(mask_reward_eval[i] +mask_action_eval[i][int(sample_mask_action[i])]- y)** 2#和eval的均方差loss=loss/self.batch_sizeprint('Loss:{}'.format(loss.cpu().item()))self.Mask_action_optimizer.zero_grad()self.Mask_Critic_optimizer.zero_grad()loss.backward()self.Mask_action_optimizer.step() self.Mask_Critic_optimizer.step()#訓(xùn)練critic和actionmask_reward0 = self.Mask_Critic_eval(sample_state[:,:,[1,2,3,5,6,7]], mask_perc_eval)loss_pg = - torch.mean(mask_reward0)#因?yàn)槭翘荻壬仙?#xff08;policy gradient，所以這里有一個(gè)負(fù)號(hào)）print('Loss_pg:{}'.format(loss_pg))self.Mask_Actor_optimizer.zero_grad()loss_pg.backward()self.Mask_Actor_optimizer.step()#訓(xùn)練actorloss_record.append([loss.cpu().item(),loss_pg.cpu().item()])#soft#對(duì)target network進(jìn)行軟更新#無論是action，actor還是critic，都需要進(jìn)行軟更新#target=（1-soft_rate)target+soft_rate*evalfor x in self.Mask_action_target.state_dict().keys():eval('self.Mask_action_target.'+x+'.data.mul_(1-self.soft_replace_rate)')eval('self.Mask_action_target.'+x+'.data.add_(self.soft_replace_rate*self.Mask_action_eval.'+x+'.data)')for x in self.Mask_Actor_target.state_dict().keys():eval('self.Mask_Actor_target.'+x+'.data.mul_((1-self.soft_replace_rate))')eval('self.Mask_Actor_target.'+x+'.data.add_(self.soft_replace_rate*self.Mask_Actor_eval.'+x+'.data)')for x in self.Mask_Critic_target.state_dict().keys():eval('self.Mask_Critic_target.'+x+'.data.mul_((1-self.soft_replace_rate))')eval('self.Mask_Critic_target.'+x+'.data.add_(self.soft_replace_rate*self.Mask_Critic_eval.'+x+'.data)')train_count += 1print('Training:epoch {}/{}\n'.format(train_count, self.train_steps))if train_count == self.train_steps:break#update the stateself.current_state=self.next_statestep_count+=1#save the modelstorch.save(self.Mask_action_eval.state_dict(), os.path.join('../model',self.city,'mask_action_model.pth'))torch.save(self.Mask_Actor_eval.state_dict(), os.path.join('../model',self.city,'mask_Actor_model.pth'))'''with open(os.path.join('../model',self.city,'loss.json'),'w') as f:json.dump(loss_record,f)'''print('Training complete!')

1.3 主函數(shù)部分

if __name__ == "__main__":os.chdir(os.path.split(os.path.realpath(__file__))[0])train_platform = RL_train(mask_total=1000000)#設(shè)置了總的口罩?jǐn)?shù)train_platform.train()

2 Net/Qnet.py

action 部分?

計(jì)算顯著性排序的DQN

import torch from torch import nn,optim import torch.nn.functional as Fclass Qnet(nn.Module):def __init__(self):super(Qnet,self).__init__()self.cov1=nn.Conv1d(in_channels=6,out_channels=16,kernel_size=5,stride=1,padding=2,bias=True)#這里池化層的輸入維度就是我環(huán)境可以看到的狀態(tài)的維度：6#分別表示：'''It（已經(jīng)感染疾病，同時(shí)被檢測(cè)出疾病的人。）Ih（被送醫(yī)治療的感染者）D（死亡人群）E（暴露人群）Iu（已經(jīng)感染疾病，但是沒有檢測(cè)的人。）R（康復(fù)人群）順序?yàn)?#xff1a;E,Iu,It,Ih,R,D'''self.cov2=nn.Conv1d(in_channels=16,out_channels=32,kernel_size=5,stride=1,padding=2,bias=True)self.cov3=nn.Conv1d(in_channels=32,out_channels=16,kernel_size=5,stride=2,padding=0,bias=True)self.cov4=nn.Conv1d(in_channels=16,out_channels=4,kernel_size=5,stride=2,padding=0,bias=True)#四層卷積層 6->16->32->16->4self.lin1=nn.Linear(in_features=664,out_features=128,bias=True)self.lin2=nn.Linear(in_features=128,out_features=7,bias=True)self.device=torch.device('cuda' if torch.cuda.is_available() else 'cpu')self.mask=torch.tensor([[0.01,0,0,0,0,0],[0,0.05,0,0,0,0],[0,0,0.01,0,0,0],[0,0,0,0.05,0,0],[0,0,0,0,0.01,0],[0,0,0,0,0,0.05]],dtype=torch.float).to(self.device)def forward(self,x):#x是[1,region_num,6]#E,Iu,It,Ih,R,Dx1=torch.matmul(x,self.mask).permute(0,2,1)#不同的狀態(tài)乘以不同的加權(quán)系數(shù)，然后將維度變成[1,6,region_num]c1=F.leaky_relu(self.cov1(x1),0.2)#[1,16,region_num]c2=F.leaky_relu(self.cov2(c1),0.2)#[1,32,region_num]c3=F.leaky_relu(self.cov3(c2),0.2)#[1,16,region_num]c4=F.leaky_relu(self.cov4(c3),0.2)#[1,4,region_num]l1=F.leaky_relu(self.lin1(c4.view(x.shape[0],-1)),0.2)#[1,4,region_num]->[1,664]([1,4*161])->[1,128]l2 = self.lin2(l1)#->[1,7]#這里是表示不同的排序原則的“打分”#論文中一共出現(xiàn)了3中不同的排序原則，其中可以任意組合，所以有2^3-1=7種（不能都不選）return l2

3 Net/Anet.py

actor 部分

import torch from torch import nn,optim import torch.nn.functional as Fclass Anet(nn.Module):def __init__(self):super(Anet,self).__init__()self.cov1=nn.Conv1d(in_channels=6,out_channels=16,kernel_size=5,stride=1,padding=2,bias=True)#這里池化層的輸入維度就是我環(huán)境可以看到的狀態(tài)的維度：6#分別表示：'''It（已經(jīng)感染疾病，同時(shí)被檢測(cè)出疾病的人。）Ih（被送醫(yī)治療的感染者）D（死亡人群）E（暴露人群）Iu（已經(jīng)感染疾病，但是沒有檢測(cè)的人。）R（康復(fù)人群）順序?yàn)?#xff1a;E,Iu,It,Ih,R,D'''self.cov2=nn.Conv1d(in_channels=16,out_channels=32,kernel_size=5,stride=1,padding=2,bias=True)self.cov3=nn.Conv1d(in_channels=32,out_channels=16,kernel_size=5,stride=2,padding=0,bias=True)self.cov4=nn.Conv1d(in_channels=16,out_channels=4,kernel_size=5,stride=2,padding=0,bias=True)#四層卷積層 6->16->32->16->4self.lin1=nn.Linear(in_features=664,out_features=128,bias=True)self.lin2=nn.Linear(in_features=128,out_features=16,bias=True)self.lin3=nn.Linear(in_features=16,out_features=1,bias=True)self.device=torch.device('cuda' if torch.cuda.is_available() else 'cpu')self.mask=torch.tensor([[0.01,0,0,0,0,0],[0,0.05,0,0,0,0],[0,0,0.01,0,0,0],[0,0,0,0.05,0,0],[0,0,0,0,0.01,0],[0,0,0,0,0,0.05]],dtype=torch.float).to(self.device)def forward(self,state):#x是[1,region_num,6]#E,Iu,It,Ih,R,Dx1 = torch.matmul(state, self.mask).permute(0, 2, 1)#不同的狀態(tài)乘以不同的加權(quán)系數(shù)，然后將維度變成[1,6,region_num]c1 = F.leaky_relu(self.cov1(x1), 0.2)#[1,16,region_num]c2 = F.leaky_relu(self.cov2(c1), 0.2)#[1,32,region_num]c3 = F.leaky_relu(self.cov3(c2), 0.2)#[1,16,region_num]c4 = F.leaky_relu(self.cov4(c3), 0.2)#[1,4,region_num]l1 = F.leaky_relu(self.lin1(c4.view(state.shape[0], -1)), 0.2)#[1,4,region_num]->[1,664]([1,4*161])->[1,128]l2 = F.leaky_relu(self.lin2(l1), 0.2)#->[1,16]l3 = torch.sigmoid(self.lin3(l2))#->[1,1]經(jīng)過sigmoid之后，相當(dāng)于是一個(gè)[0,1]之間的數(shù)return l3

4 Net/Cnet.py

critic 網(wǎng)絡(luò)

import torch from torch import nn,optim import torch.nn.functional as Fclass Cnet(nn.Module):def __init__(self,region_num):super(Cnet,self).__init__()self.lin_1 = nn.Linear(in_features=1,out_features=128,bias=True)self.lin_2 = nn.Linear(in_features=128,out_features=256,bias=True)self.lin_3 = nn.Linear(in_features=256,out_features=region_num,bias=True)self.cov1=nn.Conv1d(in_channels=7,out_channels=16,kernel_size=5,stride=1,padding=2,bias=True)self.cov2=nn.Conv1d(in_channels=16,out_channels=32,kernel_size=5,stride=1,padding=2,bias=True)self.cov3=nn.Conv1d(in_channels=32,out_channels=16,kernel_size=5,stride=2,padding=0,bias=True)self.cov4=nn.Conv1d(in_channels=16,out_channels=4,kernel_size=5,stride=2,padding=0,bias=True)self.lin1=nn.Linear(in_features=664,out_features=128,bias=True)self.lin2=nn.Linear(in_features=128,out_features=16,bias=True)self.lin3=nn.Linear(in_features=16,out_features=1,bias=True)self.device=torch.device('cuda' if torch.cuda.is_available() else 'cpu')self.mask=torch.tensor([[0.01,0,0,0,0,0],[0,0.05,0,0,0,0],[0,0,0.01,0,0,0],[0,0,0,0.05,0,0],[0,0,0,0,0.01,0],[0,0,0,0,0,0.05]],dtype=torch.float).to(self.device)def forward(self,state,perc):#state是當(dāng)前各狀態(tài)[batch_size,region_num,6]#perc是滿足因子 [batch_size,1]a1 = F.leaky_relu(self.lin_1(perc), 0.2)#[batch_size,1]——>[batch_size,128]a2 = F.leaky_relu(self.lin_2(a1), 0.2)#[batch_size,128]——>[batch_size,256]a3 = F.leaky_relu(self.lin_3(a2), 0.2)#[batch_size,256]——>[batch_size,num_region]x1 = torch.cat((torch.matmul(state, self.mask), a3.unsqueeze(2)), dim=-1).permute(0, 2, 1)#在permute之前，[batch_size,region_num,7]#[batch_size,7，region_num]c1 = F.leaky_relu(self.cov1(x1), 0.2)#[batch_size,7，region_num]->[batch_size,16，region_num]c2 = F.leaky_relu(self.cov2(c1), 0.2)#[batch_size,16，region_num]->[batch_size,32，region_num]c3 = F.leaky_relu(self.cov3(c2), 0.2)#[batch_size,32，region_num]->[batch_size,16，region_num]c4 = F.leaky_relu(self.cov4(c3), 0.2)#[batch_size,16，region_num]->[batch_size,4，region_num]l1 = F.leaky_relu(self.lin1(c4.view(state.shape[0], -1)), 0.2)#c4.view(state.shape[0], -1):[batch_size,664]#[batch_size,664]->[batch_size,128]l2 = F.leaky_relu(self.lin2(l1), 0.2)#[batch_size,128]->[batch_size,16]l3 = self.lin3(l2)#[batch_size,16]->[batch_size,1]return l3

5 simulator.py

import numpy as np import json import random import os from node import node

5.1 __init__

class simulator(object):def __init__(self,city='city_sample'):super(simulator, self).__init__()self.city = city#需要考慮的城市名稱#load dataprint('Loading dynamic pattern...')with open(os.path.join('../data',self.city,'prob.json'), 'r') as f:self.prob = np.array(json.load(f))#由于提供的代碼里面沒有這個(gè)文件，我猜測(cè)是各區(qū)域各狀態(tài)轉(zhuǎn)移比例print('Down!')print('Loading population data...')with open(os.path.join('../data',self.city,'flow.json'), 'r') as f:self.flow = np.array(json.load(f))#由于提供的代碼里面沒有這個(gè)文件，我猜測(cè)是各區(qū)域人流量with open(os.path.join('../data',self.city,'dense.json'), 'r') as f:self.dense = np.array(json.load(f))#由于提供的代碼里面沒有這個(gè)文件，我猜測(cè)是各區(qū)域人口密度with open(os.path.join('../data',self.city,'pop_region.json'), 'r') as f:self.pop = np.array(json.load(f))##由于提供的代碼里面沒有這個(gè)文件，我猜測(cè)是各區(qū)域最佳分配的數(shù)量（上限）print('Down!')self.START = 1self.region_num = len(self.flow)#區(qū)域數(shù)self.nodes = list()self.full = Falseself.parameters = [[1, 0, 0], [0, 1, 0], [0, 0, 1], [1, 1, 0], [1, 0, 1], [0, 1, 1], [1, 1, 1]]#三種排序原則，至少選擇一種，不同的排序方式（1表示選，0表示不選）

5.2 reset

def reset(self, state):# state是一個(gè)region_num*8的二維數(shù)組#初始狀態(tài)下各區(qū)域各狀態(tài)的人數(shù)print('\nReseting...')self.nodes = list()for i in range(self.region_num):self.nodes.append(node(i))#初始化id為i的點(diǎn)，初始化的時(shí)候8狀態(tài)為0self.nodes[i].set_susceptible(state[i][0])#設(shè)置i點(diǎn)易感人群數(shù)量（S）self.nodes[i].set_latent(state[i][1])#設(shè)置i點(diǎn)潛伏人群數(shù)量（E）self.nodes[i].set_infected_ut(state[i][2])#設(shè)置i點(diǎn)已經(jīng)感染，但是沒有檢測(cè)的人的數(shù)量（Iu）self.nodes[i].set_infected_t(state[i][3])#設(shè)置i點(diǎn)已經(jīng)感染疾病，但是同時(shí)被檢測(cè)出疾病的人的數(shù)量（It）self.nodes[i].set_infected_asymptomatic(state[i][4])#設(shè)置i點(diǎn)已經(jīng)感染疾病，但是沒有測(cè)出陽性的人（無癥狀感染者）的數(shù)量（Ia）self.nodes[i].set_in_hospital(state[i][5])#設(shè)置i點(diǎn)已經(jīng)感染疾病，且送醫(yī)治療的人（Ih）self.nodes[i].set_recovered(state[i][6])#設(shè)置i點(diǎn)康復(fù)人群的數(shù)量（R）self.nodes[i].set_death(state[i][7])#設(shè)置i點(diǎn)死亡人數(shù)的數(shù)量print('Down!')

5.3 simulate

各狀態(tài)轉(zhuǎn)移量模擬

def simulate(self,sim_type, #分配口罩的方案interval, #間隔多久更新一次mask_on=False, #是否戴maskmask_total=0, #整體口罩?jǐn)?shù)mask_quality=0,mask_lasting=48, #口罩有效期mask_action=0, mask_distribute_perc=[0], #口罩滿足系數(shù)bed_satisfy_perc=1,mask_satisfy_perc=1,it_move=False,is_save=False,save_time=0):print('Simulating...')#temp variablesS_temp = np.zeros((self.region_num, self.region_num + 1))L_temp = np.zeros((self.region_num, self.region_num + 1))Iut_temp = np.zeros((self.region_num, self.region_num + 1))if it_move:It_temp = np.zeros((self.region_num, self.region_num + 1))Ia_temp = np.zeros((self.region_num, self.region_num + 1))R_temp = np.zeros((self.region_num, self.region_num + 1))#不同的各狀態(tài)之間region之間的移動(dòng)比例Ih_new = 0for time in range(interval):#maskif (time % mask_lasting == 0):#如果口罩有效期過了，那么就要重新分配口罩mask_numbers = np.zeros(self.region_num)#各區(qū)域口罩分配情況if mask_on:mask_num = mask_totalif sim_type == 'Mean':mask_numbers = np.ones(self.region_num) * int(mask_num / self.region_num)#如果選擇平均分配的話，那么每個(gè)區(qū)域均分口罩?jǐn)?shù)量elif sim_type == 'Lottery':order = np.array(range(self.region_num))np.random.shuffle(order)#隨機(jī)排序區(qū)域id，按照這個(gè)排序從前到后分配口罩（排在前面的盡可能分配滿）for i in range(self.region_num):if mask_num > 0:mask_numbers[order[i]] = min(mask_num, self.pop[order[i]])#self.pop[order[i]]——該區(qū)域需要分配的數(shù)量mask_num = mask_num - mask_numbers[order[i]]else:breakelse:patient_num = np.zeros(self.region_num)patient_num_order = np.zeros(self.region_num)for i in range(self.region_num):patient_num[i] = self.nodes[i].infected_t#已經(jīng)感染疾病，但是同時(shí)被檢測(cè)出疾病的人的數(shù)量（It）patient_num_order[i] = self.nodes[i].infected_t + self.nodes[i].in_hospital + self.nodes[i].recovered + self.nodes[i].death'''已經(jīng)感染疾病，但是同時(shí)被檢測(cè)出疾病的人的數(shù)量（It）已經(jīng)感染疾病，且送醫(yī)治療的人（Ih）康復(fù)人群的數(shù)量（R）死亡人數(shù)的數(shù)量（D）這四個(gè)的和，也即累計(jì)感染人數(shù)'''if sim_type == 'Max_first':order = np.argsort(-patient_num_order)#根據(jù)累積感染人數(shù)排序（累計(jì)感染越多的越先）for i in range(self.region_num):if mask_num > 0:mask_numbers[order[i]] = min(mask_num, self.pop[order[i]])#self.pop[order[i]]——該區(qū)域需要分配的數(shù)量mask_num = mask_num - mask_numbers[order[i]]else:break#排序后，前面的區(qū)域要盡量滿足elif sim_type == 'Min_first':order = np.argsort(patient_num_order)##根據(jù)累積感染人數(shù)排序（累計(jì)感染越少的越先）for i in range(self.region_num):if mask_num > 0:mask_numbers[order[i]] = min(mask_num, self.pop[order[i]])mask_num = mask_num - mask_numbers[order[i]]else:break#排序后，前面的區(qū)域要盡量滿足elif sim_type == 'Policy':mask_parameter = self.parameters[mask_action]#mask_parameter 返回的是一個(gè)0~7的數(shù)，對(duì)應(yīng)的就是我們選擇哪種排序方式infect_num = patient_num_order/np.mean(patient_num_order)order = np.argsort(-(mask_parameter[0]*infect_num+ mask_parameter[1]*self.flow+ mask_parameter[2]*self.dense))#根據(jù)我們選擇的排序方式，決定我們是否需要感染人數(shù)、人口密度、區(qū)域人口流動(dòng)強(qiáng)度alloc_mask = np.floor(mask_num*np.array(mask_distribute_perc))#每個(gè)區(qū)域可以分配到的口罩?jǐn)?shù)量（bed_distribute_perc就是不同區(qū)域床位的滿足因子）for i in range(len(mask_distribute_perc)):mask_numbers[order[i]] = min(self.pop[order[i]], alloc_mask[i])mask_num -= mask_numbers[order[i]]#每個(gè)區(qū)域分配口罩的數(shù)量if mask_num > 0:#如果還有口罩，那么按照各區(qū)域滿足因子從大到小的順序排序，前面的優(yōu)先滿足temp_order = np.argsort(-np.array(mask_distribute_perc))for i in range(len(mask_distribute_perc)):temp=min(mask_num,self.pop[order[temp_order[i]]]-mask_numbers[order[temp_order[i]]])#排序在前面的組，按照可以滿足的口罩書填滿mask_num = mask_num - tempmask_numbers[order[temp_order[i]]] += tempif mask_num == 0:breakwhile mask_num > 0:index = np.random.randint(0, self.region_num)if mask_numbers[index] < self.pop[index]:mask_numbers[index] += 1mask_num -= 1else:break#如果還有多的口罩，那么就隨機(jī)分配elif sim_type == 'Policy_a':mask_parameter = self.parameters[mask_action]#mask_parameter 返回的是一個(gè)0~7的數(shù)，對(duì)應(yīng)的就是我們選擇哪種排序方式infect_num = patient_num_order/np.mean(patient_num_order)order = np.argsort(-(mask_parameter[0] * infect_num+ mask_parameter[1] * self.flow+ mask_parameter[2] * self.dense))#根據(jù)我們選擇的排序方式，決定我們是否需要感染人數(shù)、人口密度、區(qū)域人口流動(dòng)強(qiáng)度for i in range(self.region_num):if mask_num > 0:mask_numbers[order[i]] = min(mask_num, int(self.pop[order[i]]*mask_satisfy_perc))mask_num = mask_num - mask_numbers[order[i]]else:break#由于此時(shí)各組的滿足因子是一樣的，所以我們根據(jù)排序，從多到少分配口罩（各區(qū)域按照上限分配）#state transitionfor i in range(self.region_num):self.nodes[i].step(mask_numbers[i], mask_quality)#每一個(gè)點(diǎn)，更新各個(gè)狀態(tài)的值#cross region traveling 空間上的遷移for k in range(self.region_num):S_temp[k] = np.random.multinomial(self.nodes[k].susceptible, self.prob[((self.START-1)*48+time) % (7*48)][k])#該區(qū)域k有多少狀態(tài)為S的人會(huì)到其他區(qū)域去L_temp[k] = np.random.multinomial(self.nodes[k].latent, self.prob[((self.START-1)*48+time) % (7*48)][k])#該區(qū)域k有多少狀態(tài)為E的人會(huì)到其他區(qū)域去Iut_temp[k] = np.random.multinomial(self.nodes[k].infected_ut, self.prob[((self.START - 1)* 48 + time) % (7* 48)][k])#該區(qū)域k有多少狀態(tài)為Iu的人會(huì)到其他區(qū)域去#已經(jīng)感染疾病，但是沒有檢測(cè)的人。這些人的出行不受任何限制if it_move:It_temp[k] = np.random.multinomial(self.nodes[k].infected_t, self.prob[((self.START-1)*48+time) % (7*48)][k])#該區(qū)域k有多少狀態(tài)為It的人會(huì)到其他區(qū)域去#已經(jīng)感染疾病，同時(shí)被檢測(cè)出疾病的人。這些人的出行被限制在某一個(gè)區(qū)域內(nèi)Ia_temp[k] = np.random.multinomial(self.nodes[k].infected_asymptomatic, self.prob[((self.START - 1)* 48 + time) % (7* 48)][k])#該區(qū)域k有多少狀態(tài)為Ia的人會(huì)到其他區(qū)域去#已經(jīng)感染疾病，但是疾病監(jiān)測(cè)沒有檢測(cè)出陽性的人R_temp[k] = np.random.multinomial(self.nodes[k].recovered, self.prob[((self.START - 1)* 48 + time) % (7* 48)][k])#該區(qū)域k有多少狀態(tài)為R的人會(huì)到其他區(qū)域去S_temp_sum0 = np.sum(S_temp, axis=0)L_temp_sum0 = np.sum(L_temp, axis=0)Iut_temp_sum0 = np.sum(Iut_temp, axis=0)if it_move:It_temp_sum0 = np.sum(It_temp, axis=0)Ia_temp_sum0 = np.sum(Ia_temp, axis=0)R_temp_sum0 = np.sum(R_temp, axis=0)#別的區(qū)域過來的人數(shù)S_temp_sum1 = np.sum(S_temp, axis=1)L_temp_sum1 = np.sum(L_temp, axis=1)Iut_temp_sum1 = np.sum(Iut_temp, axis=1)if it_move:It_temp_sum1 = np.sum(It_temp, axis=1)Ia_temp_sum1 = np.sum(Ia_temp, axis=1)R_temp_sum1 = np.sum(R_temp, axis=1)#到別的區(qū)域去的人數(shù)for k in range(self.region_num):self.nodes[k].set_susceptible(self.nodes[k].susceptible+S_temp_sum0[k]-S_temp_sum1[k]+S_temp[k][self.region_num])#每個(gè)區(qū)域S狀態(tài)的變化：原先的+過來的-到別的地方去的#過來的和到別的地方去的，各用了一次S_temp[k][self.region_num]（自己到自己的），正好抵消了#所以需要再加一次self.nodes[k].set_latent(self.nodes[k].latent+L_temp_sum0[k]-L_temp_sum1[k]+L_temp[k][self.region_num])#每個(gè)區(qū)域E狀態(tài)的變化：原先的+過來的-到別的地方去的#過來的和到別的地方去的，各用了一次S_temp[k][self.region_num]（自己到自己的），正好抵消了#所以需要再加一次self.nodes[k].set_infected_ut(self.nodes[k].infected_ut+ Iut_temp_sum0[k]- Iut_temp_sum1[k]+ Iut_temp[k][self.region_num])#每個(gè)區(qū)域Iut狀態(tài)的變化：原先的+過來的-到別的地方去的#過來的和到別的地方去的，各用了一次S_temp[k][self.region_num]（自己到自己的），正好抵消了#所以需要再加一次if it_move:self.nodes[k].set_infected_t(self.nodes[k].infected_t+It_temp_sum0[k]-It_temp_sum1[k]+It_temp[k][self.region_num])#每個(gè)區(qū)域It狀態(tài)的變化：原先的+過來的-到別的地方去的#過來的和到別的地方去的，各用了一次S_temp[k][self.region_num]（自己到自己的），正好抵消了#所以需要再加一次self.nodes[k].set_infected_asymptomatic(self.nodes[k].infected_asymptomatic+Ia_temp_sum0[k]-Ia_temp_sum1[k]+Ia_temp[k][self.region_num])#每個(gè)區(qū)域Ia狀態(tài)的變化：原先的+過來的-到別的地方去的#過來的和到別的地方去的，各用了一次S_temp[k][self.region_num]（自己到自己的），正好抵消了#所以需要再加一次self.nodes[k].set_recovered(self.nodes[k].recovered+ R_temp_sum0[k]- R_temp_sum1[k]+ R_temp[k][self.region_num])#每個(gè)區(qū)域R狀態(tài)的變化：原先的+過來的-到別的地方去的#過來的和到別的地方去的，各用了一次S_temp[k][self.region_num]（自己到自己的），正好抵消了#所以需要再加一次if(is_save):save = list()for i in range(self.region_num):temp1 = [self.nodes[i].susceptible, self.nodes[i].latent, self.nodes[i].infected_ut, self.nodes[i].infected_t, self.nodes[i].infected_asymptomatic, self.nodes[i].in_hospital, self.nodes[i].recovered, self.nodes[i].death]save.append(temp1)save = np.array(save)save = save.astype(np.float)with open(os.path.join('../result',self.city,'result_'+str(save_time*interval+time)+'.json'), 'w') as f:json.dump(save.tolist(), f)next_state = list()for i in range(self.region_num):next_state.append([self.nodes[i].susceptible,self.nodes[i].latent,self.nodes[i].infected_ut,self.nodes[i].infected_t,self.nodes[i].infected_asymptomatic,self.nodes[i].in_hospital,self.nodes[i].recovered,self.nodes[i].death])#每個(gè)區(qū)域的狀態(tài)a = self.statistic()print('S:{},L:{},Iut:{},It:{},Ia:{},Ih:{},R:{},D:{}'.format(a[0], a[1], a[2], a[3], a[4], a[5], a[6], a[7]))print('Is Full:{}'.format(self.full))return np.array(next_state), Ih_new

6 node.py

import numpy as np Pa = 0.018 # false negative rate eps_1 = 250 # average length of incubation (30min) eps_1_d = 5.2 # average length of incubation (day) d = 0.15 # death rate t = 672 # average recovery time (30min) t_d = 14 # average recovery time (day) d_hospital = 0.04 # death rate in hospital t_hospital = 614 # average recovery time in hospital(30min) t_hospital_d = 12.8 # average recovery time in hospital(day) R_0 = 2.68 # basic reproduction number r_a = 0.6 r_L = 1.0 eps = 1/eps_1 beta = R_0 / (r_L * eps_1_d + (Pa * r_a + (1 - Pa)) * t_d) beta = np.power(1+beta, 1/48)-1 L_I = eps * (1 - Pa) L_Ia = eps * Pa I_D = d / t I_R = ((1 - d)/(t_d - d))/48theta = 48 # testing speed (30min) I_h = 1/theta Ia_R = 1 / t Ih_D = d_hospital / t_hospital Ih_R = ((1 - d_hospital) / (t_hospital_d - d_hospital))/48class node:def __init__(self, id):self.id = idself.susceptible = 0self.infected_ut = 0self.infected_t = 0self.death = 0self.in_hospital = 0self.infected_asymptomatic = 0self.recovered = 0self.latent = 0def set_susceptible(self, susceptible):self.susceptible = susceptibledef set_latent(self, latent):self.latent = latentdef set_infected_ut(self, infected_ut):self.infected_ut = infected_utdef set_infected_t(self, infected_t):self.infected_t = infected_tdef set_infected_asymptomatic(self, infected_asymptomatic):self.infected_asymptomatic = infected_asymptomaticdef set_death(self, death):self.death = deathdef set_in_hospital(self, in_hospital):self.in_hospital = in_hospitaldef set_recovered(self, recovered):self.recovered = recovereddef step(self,mask_number,mask_quality):if (self.susceptible + self.latent + self.infected_ut + self.infected_t + self.infected_asymptomatic + self.in_hospital + self.recovered > 0):#除去了死亡狀態(tài)，其他狀態(tài)人數(shù)和大于0（換句話說，就是人沒有全死光）mask_factor = 1 - np.clip(mask_number / (self.susceptible +self.latent +self.infected_ut +self.infected_t +self.infected_asymptomatic +self.in_hospital +self.recovered),0, 1) * mask_quality#除的這一串相當(dāng)于是這個(gè)node目前有的口罩?jǐn)?shù)/目前存活的人#也即論文中說的口罩覆蓋率Π(t)#也就是這一串是論文中的(1-Π(t)γ)lambda_j = ((self.infected_ut +self.infected_t +self.infected_asymptomatic * r_a +self.latent * r_L) /(self.susceptible +self.latent +self.infected_ut + self.infected_t +self.infected_asymptomatic +self.in_hospital +self.recovered)) * beta * mask_factor#前面除的那一串是該node感染人數(shù)/該node總?cè)藬?shù) * β * (1-Π(t)γ)susceptible_to_latent, __ = np.random.multinomial(self.susceptible, [lambda_j, 1])#相當(dāng)于是從S 狀態(tài)到E 狀態(tài)的人數(shù)#我們?cè)纫还灿衧elf.susceptible個(gè)S狀態(tài)的人，每個(gè)人都有l(wèi)ambda_j的概率變成E狀態(tài)#于是我們就用多次二項(xiàng)分布的方法，判斷每一個(gè)個(gè)體是否會(huì)變成E狀態(tài)self.susceptible -= susceptible_to_latentself.latent += susceptible_to_latent#S狀態(tài)增加susceptible_to_latent個(gè)人，E狀態(tài)減少susceptible_to_latent個(gè)人latent_to_infected, latent_to_Ia, __ = np.random.multinomial(self.latent, [L_I, L_Ia, 1])self.infected_ut += latent_to_infectedself.infected_asymptomatic += latent_to_Iaself.latent -= (latent_to_Ia + latent_to_infected)#從E轉(zhuǎn)換到Iu和Ia的人數(shù)prob = I_hinfected_ut_to_t, __ = np.random.multinomial(self.infected_ut, [prob, 1])self.infected_ut -= infected_ut_to_tself.infected_t += infected_ut_to_t#從Iu轉(zhuǎn)換到Ia的人數(shù)infected_to_death, infected_to_recovered, __ = np.random.multinomial(self.infected_ut, [I_D, I_R, 1])self.death += infected_to_deathself.recovered += infected_to_recoveredself.infected_ut -= (infected_to_death + infected_to_recovered)#從Iu狀態(tài)（已經(jīng)感染疾病，但是沒有檢測(cè)的人）到死亡/康復(fù)的人infected_to_death, infected_to_recovered, __ = np.random.multinomial(self.infected_t, [I_D, I_R, 1])self.death += infected_to_deathself.recovered += infected_to_recoveredself.infected_t -= (infected_to_death + infected_to_recovered)#從It狀態(tài)（已經(jīng)感染疾病，同時(shí)被檢測(cè)出疾病的人）到死亡/康復(fù)的人Ia_to_recovered, __ = np.random.multinomial(self.infected_asymptomatic, [Ia_R, 1])self.recovered += Ia_to_recoveredself.infected_asymptomatic -= Ia_to_recovered#從Ia狀態(tài)（已經(jīng)感染疾病，但是疾病監(jiān)測(cè)沒有檢測(cè)出陽性的人）【無癥狀感染者】到康復(fù)的人數(shù)in_hospital_to_death, in_hospital_to_recovered, __ = np.random.multinomial(self.in_hospital, [Ih_D, Ih_R, 1])self.death += in_hospital_to_deathself.recovered += in_hospital_to_recoveredself.in_hospital -= (in_hospital_to_death + in_hospital_to_recovered)#從Ih狀態(tài)（送醫(yī)治療的人）到康復(fù)/死亡的人

總結(jié)

以上是生活随笔為你收集整理的论文代码解读 Hierarchical Reinforcement Learning for Scarce Medical Resource Allocation的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。