-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdoom.py
250 lines (214 loc) · 8.62 KB
/
doom.py
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
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
from vizdoom import *
import torch
import torchvision
import torchvision.transforms as transforms
import matplotlib.pyplot as plt
from torch.autograd import Variable
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from collections import deque
import random
import time
import copy
import argparse
import cv2
import warnings
warnings.filterwarnings('ignore')
from model import DDDQNNet , init_weights
from utils import SumTree , Memory , predict_action ,create_enviornment , preprocess_frame
parser = argparse.ArgumentParser(description="Agent DOoM")
parser.add_argument('--learning_rate', type=float,help='Learning rate for DQN',default=0.025)
parser.add_argument('--episodes', type=int,help='Number of episodes for training',default=5000)
parser.add_argument('--steps', type=int,help='Maximum steps per episode',default=5000)
parser.add_argument('--tau', type=int,help='Time to refil target network weights',default=3000)
parser.add_argument('--explore_stop', type=float,help='Stopping exploration probability',default=0.01)
parser.add_argument('--decay_rate', type=float,help='Decay rate for exploration probability',default=0.00005)
parser.add_argument('--discount_factor', type=float,help='Discounting factor for returns',default=0.95)
parser.add_argument('--pretrain_length', type=int,help='number of iterations to fill replay memory',default=10000)
parser.add_argument('--mem_size', type=int,help='Size of the replay memory',default=10000)
parser.add_argument('--train', type=bool,help='want to train or not',default=False)
parser.add_argument('--batch_size', type=int,help='Batch size to sample from replay memory',default=64)
parser.add_argument('--gpu', type=bool,help='Use GPU',default=False)
parser.add_argument('--weights', type=str,help='load pretrained weights',default=None)
parser.add_argument('--save_weights', type=bool,help='save the DQN weights',default=False)
parser.add_argument('--graph', type=bool,help='save the DQN weights',default=False)
parser.add_argument('--test', type=bool,help='Generate testing video',default=False)
args = parser.parse_args()
game,possible_actions = create_enviornment()
stack_size=4
stacked_frames = deque([np.zeros((108,124),dtype=np.int) for i in range(stack_size)],maxlen=4)
def stack_frames(stacked_frames,state,is_new_episode):
frame = preprocess_frame(state)
if is_new_episode:
stacked_frames = deque([frame for i in range(stack_size)],maxlen=4)
else:
stacked_frames.append(frame)
stacked_state = torch.stack(list(stacked_frames),1)
return stacked_state, stacked_frames
state_size = [4,108,124]
action_size = game.get_available_buttons_size()
explore_start = 1.0
episode_render = False
if args.weights is None:
DQNetwork = DDDQNNet(state_size,action_size,name="DQNetwork")
init_weights(DQNetwork)
else:
DQNetwork = torch.load(args.weights)
print("Weights loaded")
TargetNetwork = DDDQNNet(state_size,action_size,name="TargetNetwork")
if args.gpu:
DQNetwork , TargetNetwork = DQNetwork.cuda() , TargetNetwork.cuda()
optimizerDQ = optim.RMSprop(DQNetwork.parameters(),lr=args.learning_rate)
memory = Memory(args.mem_size)
game.new_episode()
if args.train:
print("Training started")
for i in range(args.pretrain_length):
if i==0:
state = game.get_state().screen_buffer
state,stacked_frames = stack_frames(stacked_frames,state,True)
action=random.choice(possible_actions)
reward = game.make_action(action)
done = game.is_episode_finished()
if done:
next_state = np.zeros(state.shape,dtype=np.float32)
experience = state,action,reward,next_state,done
memory.store(experience)
game.new_episode()
state = game.get_state().screen_buffer
state,stacked_frames = stack_frames(stacked_frames,state,True)
else:
next_state = game.get_state().screen_buffer
next_state,stacked_frames = stack_frames(stacked_frames,next_state,False)
experience = state,action,reward,next_state,done
memory.store(experience)
state= next_state
Summary = { "Loss":[] , "Rewards":[] }
decay_step=0
tau=0
game.init()
TargetNetwork = copy.deepcopy(DQNetwork)
for episode in range(args.episodes):
loss=0
step=0
episode_rewards=[]
game.new_episode()
state = game.get_state().screen_buffer
state,stacked_frames = stack_frames(stacked_frames,state,True)
while step<args.steps:
step+=1
tau+=1
decay_step+=1
action,explore_probability,is_greedy= predict_action(explore_start,args.explore_stop,args.decay_rate,decay_step,state,possible_actions,DQNetwork,args.gpu)
reward = game.make_action(action)
done = game.is_episode_finished()
episode_rewards.append(reward)
optimizerDQ.zero_grad()
if done:
next_state=np.zeros(state.shape,dtype=np.float32)
step=args.steps
total_reward = np.sum(episode_rewards)
Summary["Rewards"].append(total_reward)
Summary["Loss"].append(loss)
print('Episode: {}'.format(episode),'Total reward: {}'.format(total_reward),
'Training loss: {:.4f}'.format(loss),
'Explore P: {:.4f}'.format(explore_probability),'Is Greedy: {}'.format(is_greedy))
experience= state,action,reward,next_state,done
memory.store(experience)
else:
next_state = game.get_state().screen_buffer
next_state , stacked_frames = stack_frames(stacked_frames,next_state,False)
experience= state,action,reward,next_state,done
memory.store(experience)
state = next_state
ISWeights_mb,tree_idx,batch = memory.sample(args.batch_size)
ISWeights_mb = torch.tensor(ISWeights_mb,requires_grad=True)
if args.gpu:
ISWeights_mb=ISWeights_mb.cuda()
ISWeights_mb = ISWeights_mb.squeeze(1)
states_mb = torch.cat([Variable(each[0][0],requires_grad=True) for each in batch],0)
actions_mb = torch.cat([torch.tensor(each[0][1],requires_grad=True) for each in batch],0)
rewards_mb = torch.tensor([each[0][2] for each in batch],requires_grad=True)
next_states_mb = torch.cat([torch.tensor(each[0][3],requires_grad=True) for each in batch], 0)
dones_mb = [each[0][4] for each in batch]
target_Qs_batch =[]
predicted_Qs_batch=[]
if args.gpu:
states_mb,next_states_mb = states_mb.cuda() , next_states_mb.cuda()
q_state = DQNetwork(states_mb)
q_next_state = DQNetwork(next_states_mb)
q_target_next_state = TargetNetwork(next_states_mb)
for i in range(len(batch)):
terminal = dones_mb[i]
_,action = torch.max(q_next_state[i],0)
_,ac = torch.max(actions_mb[i],0)
predicted_Qs_batch.append(q_state[i][ac])
if terminal:
target_Qs_batch.append(rewards_mb[i])
else:
target = rewards_mb[i].cuda() + args.discount_factor*q_target_next_state[i][action]
target_Qs_batch.append(target)
targets_mb = torch.tensor([each for each in target_Qs_batch],requires_grad=True)
pre_targets_mb = torch.tensor([each for each in predicted_Qs_batch],requires_grad=True)
if args.gpu:
targets_mb , pre_targets_mb = targets_mb.cuda() , pre_targets_mb.cuda()
absolute_errors = torch.abs(targets_mb-pre_targets_mb)
loss = (targets_mb-pre_targets_mb)*(targets_mb-pre_targets_mb)
loss = torch.mean(ISWeights_mb*loss)
loss = Variable(loss,requires_grad=True)
loss.backward()
optimizerDQ.step()
if args.gpu:
memory.batch_update(tree_idx,absolute_errors.cpu().detach().numpy())
else:
memory.batch_update(tree_idx,absolute_errors.detach().numpy())
if tau>args.tau:
TargetNetwork = copy.deepcopy(DQNetwork)
tau=0
print("Model updated")
if episode%5==0 and args.save_weights:
torch.save(DQNetwork,'./dqnet.pt')
print("Model Saved")
if args.graph:
x = [i+1 for i in range(len(Summary["Loss"]))]
y1 = Summary["Loss"]
y2 = Summary["Rewards"]
plt.subplot(2,1,1)
plt.plot(x,y1)
plt.title("Summary Graphs")
plt.ylabel("Loss")
plt.subplot(2,1,2)
plt.plot(x,y2)
plt.xlabel("episodes")
plt.ylabel("Rewads")
plt.savefig('./graph.png')
if args.test:
print("Testing")
for i in range(10):
fourcc = cv2.VideoWriter_fourcc(*'DIVX')
video = cv2.VideoWriter('./video{}.avi'.format(i),fourcc,10,(320,240))
game.new_episode()
state = game.get_state().screen_buffer
st = state.transpose(1,2,0)
video.write(st)
state,stacked_frames = stack_frames(stacked_frames,state,True)
s=0
while not game.is_episode_finished():
s+=1
action,explore_probability,is_greedy = predict_action(0.01,0,0,0,state,possible_actions,DQNetwork,args.gpu)
game.make_action(action)
done = game.is_episode_finished()
if done:
break
else:
next_state = game.get_state().screen_buffer
st = next_state.transpose(1,2,0)
video.write(st)
next_state,stacked_frames = stack_frames(stacked_frames,next_state,False)
state = next_state
score = game.get_total_reward()
print("Score : ",score,"Steps :",s)
video.release()
game.close()