forked from ericdanz/ReinforcingRobots
-
Notifications
You must be signed in to change notification settings - Fork 0
/
make_phong_states.py
250 lines (219 loc) · 10.1 KB
/
make_phong_states.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
import numpy as np
import copy
import cv2
from phong_simulator import Simulator
import time
from tqdm import tqdm
def make_states(simulator,actor,epsilon,number_of_steps,number_of_games,winners_only=False,play_itself=1):
#initialize something to hold the games
#assume epsilon decay happens outside this function
game_list = []
last_score_list = []
for i in tqdm(range(number_of_games)):
state_list = make_one_set(simulator,actor,epsilon,number_of_steps,play_itself=play_itself)
last_score_list.append(state_list[-1][0][1]) #change this to average score
game_list = game_list + state_list
return game_list,np.mean(last_score_list)
def make_one_set(simulator,actor,epsilon,number_of_steps,display=False,play_itself=1):
state_list = []
left_state_list = []
previous_state = np.zeros((simulator.screen_size,simulator.screen_size,3))
last_score = 0
action_numerator = 0
per_action_penalty = 0
simulator.reset()
previous_state[:,:,0] = np.reshape(simulator.screen,(simulator.screen_size,simulator.screen_size))
cv2.destroyAllWindows()
#one set is one score ?
for i in range(number_of_steps):
state = []
#check if the ball has crossed to the left field
if simulator.ball_side() == "left" and play_itself:
#flip the screen and play for the right side
# screen = simulator.screen[:,::-1]
#now do actions, e-greedy
if np.random.uniform() < epsilon:
#do a random action
left_action = np.random.randint(actor.number_of_actions)
else:
left_action = actor.return_action(previous_state[:,::-1])
simulator.do_action(left_action,side="left")
elif simulator.ball_side() == "left":
#do pong AI
simulator.do_action(None,side="left",simple_AI=True)
left_action = 0
else:
left_action = 0
screen = simulator.screen
#now do actions, e-greedy
if np.random.uniform() < epsilon:
#do a random action
action = np.random.randint(actor.number_of_actions)
else:
action = actor.return_action(previous_state)
screen,score,points_made,end = simulator.do_action(action,side="right")
if action != 0:
action_numerator += 1
# if end == 0:
# #move the game along with a no-op!
# screen,score,points_made,end = simulator.do_action(0,side="right")
if display:
print('no-op ', 'up ','down')
print(actor.display_output)
if play_itself:
display_title = "Phong RL vs RL"
else:
display_title = "Phong Simple vs RL"
padded_screen = np.ones((screen.shape[0]+10,screen.shape[1]+10))
padded_screen[5:screen.shape[0]+5,5:screen.shape[1]+5] = screen[:,:,0]
cv2.imshow(display_title,cv2.resize(padded_screen,(0,0),fx=6,fy=6))
cv2.moveWindow(display_title,10,10)
cv2.waitKey(100)
state.append([screen,score,action,points_made,previous_state])
state_list.append(copy.deepcopy(state))
state.append([screen[:,::-1],score,left_action,points_made,previous_state[:,::-1]])
left_state_list.append(copy.deepcopy(state))
previous_state[:,:,1:] = np.copy(previous_state[:,:,:2])
previous_state[:,:,0] = np.reshape(screen,(simulator.screen_size,simulator.screen_size))
if end != 0:
#propogate the score backwards
#figure out the direction of the score
reward = simulator.reward*end #-1 if point on the actor, 1 otherwise
discount_iterator = 0
discount_factor = 0.99
action_penalty = -.1
if action_numerator !=0:
per_action_penalty = action_penalty/action_numerator
#actually, lets table the action penalty for now
per_action_penalty = 0
for previous_state in reversed(state_list):
#this starts at the win, but doesn't add to the win reward
#the reward can be decreased linearly or exponentially
#this will do it linearly
if previous_state[0][2] != 0:
previous_state[0][1] += per_action_penalty
previous_state[0][1] += reward*np.power(discount_factor,discount_iterator)
discount_iterator += 1
# #do the same, backwards, for left_action
# for previous_state in reversed(left_state_list):
# #this starts at the win, but doesn't add to the win reward
# #the reward can be decreased linearly or exponentially
# #this will do it linearly
# previous_state[0][1] += -1*reward*np.power(discount_factor,discount_iterator)
# discount_iterator += 1
simulator.reset()
break
return state_list #+ left_state_list
def run_test_games(simulator,actor,number_of_steps,display=True,play_itself=0):
state_list = []
score_list = []
left_state_list = []
previous_state = np.zeros((simulator.screen_size,simulator.screen_size,3))
last_score = 0
action_numerator = 0
per_action_penalty = 0
epsilon=0
simulator.reset()
previous_state[:,:,0] = np.reshape(simulator.screen,(simulator.screen_size,simulator.screen_size))
cv2.destroyAllWindows()
#one set is one score ?
for i in range(number_of_steps):
state = []
#check if the ball has crossed to the left field
if simulator.ball_side() == "left" and play_itself:
#flip the screen and play for the right side
# screen = simulator.screen[:,::-1]
#now do actions, e-greedy
if np.random.uniform() < epsilon:
#do a random action
left_action = np.random.randint(actor.number_of_actions)
else:
left_action = actor.return_action(previous_state[:,::-1])
simulator.do_action(left_action,side="left")
elif simulator.ball_side() == "left":
#do pong AI
simulator.do_action(None,side="left",simple_AI=True)
left_action = 0
else:
left_action = 0
screen = simulator.screen
#now do actions, e-greedy
if np.random.uniform() < epsilon:
#do a random action
action = np.random.randint(actor.number_of_actions)
else:
action = actor.return_action(previous_state)
screen,score,points_made,end = simulator.do_action(action,side="right")
if action != 0:
action_numerator += 1
# if end == 0:
# #move the game along with a no-op!
# screen,score,points_made,end = simulator.do_action(0,side="right")
if display:
# print('no-op ', 'up ','down')
# print(actor.display_output)
if play_itself:
display_title = "Phong RL vs RL"
else:
display_title = "Phong Simple vs RL"
padded_screen = np.ones((screen.shape[0]+10,screen.shape[1]+10))
padded_screen[5:screen.shape[0]+5,5:screen.shape[1]+5] = screen[:,:,0]
cv2.imshow(display_title,cv2.resize(padded_screen,(0,0),fx=6,fy=6))
cv2.moveWindow(display_title,10,10)
cv2.waitKey(70)
state.append([screen,score,action,points_made,previous_state])
state_list.append(copy.deepcopy(state))
score_list.append(copy.copy(score))
# state.append([screen[:,::-1],score,left_action,points_made,previous_state[:,::-1]])
# left_state_list.append(copy.deepcopy(state))
previous_state[:,:,1:] = np.copy(previous_state[:,:,:2])
previous_state[:,:,0] = np.reshape(screen,(simulator.screen_size,simulator.screen_size))
# if end != 0:
# #propogate the score backwards
# #figure out the direction of the score
# reward = simulator.reward*end #-1 if point on the actor, 1 otherwise
# discount_iterator = 0
# discount_factor = 0.99
#
# action_penalty = -.1
# if action_numerator !=0:
# per_action_penalty = action_penalty/action_numerator
# #actually, lets table the action penalty for now
# per_action_penalty = 0
#
# for previous_state in reversed(state_list):
# #this starts at the win, but doesn't add to the win reward
# #the reward can be decreased linearly or exponentially
# #this will do it linearly
# if previous_state[0][2] != 0:
# previous_state[0][1] += per_action_penalty
# previous_state[0][1] += reward*np.power(discount_factor,discount_iterator)
# discount_iterator += 1
# # #do the same, backwards, for left_action
# # for previous_state in reversed(left_state_list):
# # #this starts at the win, but doesn't add to the win reward
# # #the reward can be decreased linearly or exponentially
# # #this will do it linearly
# # previous_state[0][1] += -1*reward*np.power(discount_factor,discount_iterator)
# # discount_iterator += 1
#
# break
return score_list #+ left_state_list
class FakeActor:
def __init__(self,num_actions):
self.number_of_actions = num_actions
self.display_output = [0,0]
def return_action(self,simulator_screen):
return np.random.randint(self.number_of_actions)
if __name__ == "__main__":
sim = Simulator(10)
actor = FakeActor(3)
start_time = time.time()
game_state_list = make_states(sim,actor,1,200,10,winners_only=False)
print(time.time() - start_time)
for state_list in game_state_list:
for state in state_list:
print(state[0][1:4],np.mean(state[0][4]))
cv2.imshow('Phong!',cv2.resize(state[0][4],(0,0),fx=2,fy=2))
np.save('phong_screen',state[0][4])
cv2.waitKey(400)