-
Notifications
You must be signed in to change notification settings - Fork 0
/
dqn.py
232 lines (182 loc) · 8.43 KB
/
dqn.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
import cv2
import numpy as np
from collections import deque
import random
import torch
import torch.nn as nn
import torch.optim as optim
import torchvision.transforms as T
import torch.nn.functional as F
from gameTRY import Breakout
import os
import sys
import time
class NeuralNetwork(nn.Module):
def __init__(self):
super(NeuralNetwork, self).__init__()
self.number_of_actions = 2
self.gamma = 0.99
self.final_epsilon = 0.05
self.initial_epsilon = 0.1
self.number_of_iterations = 2000000
self.replay_memory_size = 750000
self.minibatch_size = 32
self.explore = 3000000 # Timesteps to go from INITIAL_EPSILON to FINAL_EPSILON
self.conv1 = nn.Conv2d(4, 32, kernel_size = 8, stride = 4)
self.conv2 = nn.Conv2d(32, 64, 4, 2)
self.conv3 = nn.Conv2d(64, 64, 3, 1)
self.fc4 = nn.Linear(7 * 7 * 64, 512)
self.fc5 = nn.Linear(512, self.number_of_actions)
def forward(self, x):
x = F.relu(self.conv1(x))
x = F.relu(self.conv2(x))
x = F.relu(self.conv3(x))
#make sure input tensor is flattened
x = F.relu(self.fc4(x.view(x.size(0), -1)))
return self.fc5(x)
def preprocessing(image):
image_data = cv2.cvtColor(cv2.resize(image, (84, 84)), cv2.COLOR_BGR2GRAY)
image_data[image_data > 0] = 255
image_data = np.reshape(image_data,(84, 84, 1))
image_tensor = image_data.transpose(2, 0, 1)
image_tensor = image_tensor.astype(np.float32)
image_tensor = torch.from_numpy(image_tensor)
return image_tensor
def init_weights(m):
if type(m) == nn.Conv2d or type(m) == nn.Linear:
torch.nn.init.uniform(m.weight, -0.01, 0.01)
m.bias.data.fill_(0.01)
def train(model, start):
# define Adam optimizer
optimizer = optim.Adam(model.parameters(), lr=0.0002)
# initialize mean squared error loss
criterion = nn.MSELoss() # crossentropy
# instantiate game
game_state = Breakout()
# initialize replay memory
D = deque()
#replay = []
# initial action is do nothing
action = torch.zeros([model.number_of_actions], dtype=torch.float32)
action[0] = 0
image_data, reward, terminal = game_state.take_action(action)
image_data = preprocessing(image_data)
state = torch.cat((image_data, image_data, image_data, image_data)).unsqueeze(0) # 1-4-84-84
# initialize epsilon value
epsilon = model.initial_epsilon
iteration = 0
#epsilon = 0.0927
#iteration = 420000
# main infinite loop
while iteration < model.number_of_iterations:
# get output from the neural network
output = model(state)[0] # Output size = torch.Size([2]) tensor([-0.0278, 1.7244]
#output = model(state)
# initialize action
action = torch.zeros([model.number_of_actions], dtype=torch.float32)
# epsilon greedy exploration
random_action = random.random() <= epsilon
if random_action:
print("Random action!")
# Pick action --> random or index of maximum q value
action_index = [torch.randint(model.number_of_actions, torch.Size([]), dtype=torch.int)
if random_action
else torch.argmax(output)][0]
#print("Action index shape: ", action_index.shape) # torch.Size([])
action[action_index] = 1
if epsilon > model.final_epsilon:
epsilon -= (model.initial_epsilon - model.final_epsilon) / model.explore
# get next state and reward
image_data_1, reward, terminal = game_state.take_action(action)
image_data_1 = preprocessing(image_data_1)
#print("İmage data_1 shape: ", image_data_1.shape) # 1-84-84
state_1 = torch.cat((state.squeeze(0)[1:, :, :], image_data_1)).unsqueeze(0) # squeeze(0).shape = 4-84-84
#print("State_1 Shape: ", state_1.shape) # State_1 Shape = ([1, 4, 84, 84]) # squeeze(0)[1:,:,:].shape = 3-84-84
action = action.unsqueeze(0)
#print("Action size: ", action.shape) # 1-2
reward = torch.from_numpy(np.array([reward], dtype=np.float32)).unsqueeze(0)
#print("Reward size: ", reward.shape)
# save transition to replay memory
D.append((state, action, reward, state_1, terminal))
# if replay memory is full, remove the oldest transition
if len(D) > model.replay_memory_size:
D.popleft()
# sample random minibatch
# it picks k unique random elements, a sample, from a sequence: random.sample(population, k)
minibatch = random.sample(D, min(len(D), model.minibatch_size))
# unpack minibatch
state_batch = torch.cat(tuple(d[0] for d in minibatch))
#print("state_batch size: ", state_batch.shape)
action_batch = torch.cat(tuple(d[1] for d in minibatch))
#print("action_batch size: ", action_batch.shape)
reward_batch = torch.cat(tuple(d[2] for d in minibatch))
#print("reward_batch size: ", reward_batch.shape)
state_1_batch = torch.cat(tuple(d[3] for d in minibatch))
#print("state_1_batch size: ", state_1_batch.shape)
# get output for the next state
output_1_batch = model(state_1_batch)
#print("output_1_batch: " , output_1_batch.shape)
# set y_j to r_j for terminal state, otherwise to r_j + gamma*max(Q) Target Q value Bellman equation.
y_batch = torch.cat(tuple(reward_batch[i] if minibatch[i][4]
else reward_batch[i] + model.gamma * torch.max(output_1_batch[i])
for i in range(len(minibatch))))
# extract Q-value -----> column1 * column1 + column2 * column2
# The main idea behind Q-learning is that if we had a function Q∗ :State × Action → ℝ
#that could tell us what our return would be, if we were to take an action in a given state,
#then we could easily construct a policy that maximizes our rewards
q_value = torch.sum(model(state_batch) * action_batch, dim=1)
#print("q_value: ", q_value.shape)
# PyTorch accumulates gradients by default, so they need to be reset in each pass
optimizer.zero_grad()
# returns a new Tensor, detached from the current graph, the result will never require gradient
y_batch = y_batch.detach()
# calculate loss
loss = criterion(q_value, y_batch)
# do backward pass
loss.backward()
optimizer.step()
# set state to be state_1
state = state_1
iteration += 1
if iteration % 10000 == 0:
torch.save(model, "trained_model/current_model_" + str(iteration) + ".pth")
print("total iteration: {} Elapsed time: {:.2f} epsilon: {:.5f}"
" action: {} Reward: {:.1f}".format(iteration,((time.time() - start)/60),epsilon,action_index.cpu().detach().numpy(),reward.numpy()[0][0]))
def test(model):
game_state = Breakout()
# initial action is do nothing
action = torch.zeros([model.number_of_actions], dtype=torch.float32)
action[0] = 1
image_data, reward, terminal = game_state.take_action(action)
image_data = preprocessing(image_data)
state = torch.cat((image_data, image_data, image_data, image_data)).unsqueeze(0)
while True:
# get output from the neural network
output = model(state)[0]
action = torch.zeros([model.number_of_actions], dtype=torch.float32)
# get action
action_index = torch.argmax(output)
action[action_index] = 1
# get next state
image_data_1, reward, terminal = game_state.take_action(action)
image_data_1 = preprocessing(image_data_1)
state_1 = torch.cat((state.squeeze(0)[1:, :, :], image_data_1)).unsqueeze(0)
# set state to be state_1
state = state_1
def main(mode):
if mode == 'test':
model = torch.load('trained_model/current_model_420000.pth', map_location='cpu').eval()
test(model)
elif mode == 'train':
if not os.path.exists('trained_model/'):
os.mkdir('trained_model/')
model = NeuralNetwork()
model.apply(init_weights)
start = time.time()
train(model, start)
elif mode == 'continue': # You can change trained model id and keep training.
model = torch.load('trained_model/current_model_420000.pth', map_location='cpu').eval()
start = time.time()
train(model, start)
if __name__ == "__main__":
main(sys.argv[1])