-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathactor.py
383 lines (319 loc) · 14.6 KB
/
actor.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
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
import zmq
import tensorflow as tf
import numpy as np
import collections
import cv2
import gym
import minerl
import cv2
import matplotlib.pyplot as plt
import argparse
from absl import flags
from absl import logging
import os
from utils import TrajectoryInformation, DummyDataLoader, TrajectoryDataPipeline, create_nodes
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" # see issue #152
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
parser = argparse.ArgumentParser(description='MineRL IMPALA Actor')
parser.add_argument('--env_id', type=int, default=0, help='ID of environment')
arguments = parser.parse_args()
writer = tf.summary.create_file_writer("tensorboard")
context = zmq.Context()
# Socket to talk to server
print("Connecting to hello world server…")
socket = context.socket(zmq.REQ)
socket.connect("tcp://localhost:" + str(5555 + arguments.env_id))
data_path = '/home/kimbring2/minerl_data/'
env_name = 'MineRLObtainDiamondDense-v0'
trajectory_name = 'v3_absolute_grape_changeling-6_37339-46767'
trajectory_information = TrajectoryInformation(data_path + env_name + "/" + trajectory_name)
final_chain = trajectory_information.to_old_chain_format(items=trajectory_information.chain, return_time_indexes=False)
nodes = create_nodes(final_chain)
#for node in nodes:
# print("node.name: ", node.name)
item_list = ['log', 'planks', 'crafting_table', 'stick', 'wooden_pickaxe', 'cobblestone', 'coal',
'stone_pickaxe', 'iron_ore', 'furnace', 'iron_ingot', 'iron_pickaxe', 'stone']
#env = gym.make('MineRLNavigateDense-v0')
#env = gym.make('MineRLObtainDiamondDense-v0')
env = gym.make('MineRLTreechop-v0')
num_actions = 43
state_size = (64,64,3)
scores = []
episodes = []
average = []
for episode_step in range(0, 2000000):
observation = env.reset()
node_index = 0
if nodes[node_index].name in item_list:
item_index = item_list.index(nodes[node_index].name)
else:
item_index = 0
scale = len(item_list)
item_layer = np.zeros([64,64,scale], dtype=np.float32)
item_layer[:, :, item_index] = 1.0
inventory_channel = np.zeros((64,64,1))
if 'inventory' in observation:
region_max_height = observation['pov'].shape[0]
region_max_width = observation['pov'].shape[1]
rs = 8
if min(region_max_height, region_max_width) < rs:
raise ValueError("'region_size' is too large.")
num_element_width = region_max_width // rs
inventory_channel = np.zeros(shape=list(observation['pov'].shape[:-1]) + [1],
dtype=observation['pov'].dtype)
for key_idx, key in enumerate(observation['inventory'].keys()):
item_scaled = np.clip(1 - 1 / (observation['inventory'][key] + 1), # Inversed
0, 1)
item_channel = np.ones(shape=[rs, rs, 1], dtype=observation['pov'].dtype) * item_scaled
width_low = (key_idx % num_element_width) * rs
height_low = (key_idx // num_element_width) * rs
if height_low + rs > region_max_height:
raise ValueError("Too many elements on 'inventory'. Please decrease 'region_size' of each component.")
inventory_channel[height_low:(height_low + rs), width_low:(width_low + rs), :] = item_channel
state = np.concatenate((observation['pov'] / 255.0, inventory_channel), axis=2)
state = np.concatenate((state, item_layer), axis=2)
state = tf.constant(state, dtype=tf.float32)
state = observation['pov'] / 255.0
'''
pov_array = observation['pov'] / 255.0
compassAngle_array = observation['compassAngle'] / 360.0
compassAngle_array = np.ones((64,64,1)) * compassAngle_array
state = np.concatenate((pov_array, compassAngle_array), 2)
state = np.expand_dims(state, 0)
'''
done = False
reward = 0.0
reward_sum = 0
step = 0
while True:
try:
state_reshaped = np.reshape(state, (1,*state_size))
env_output = {"env_id": np.array([arguments.env_id]),
"reward": reward / 10.0,
"done": done,
"observation": state_reshaped}
socket.send_pyobj(env_output)
action_index = int(socket.recv_pyobj()['action'])
print("action_index:" , action_index)
action = env.action_space.noop()
'''
if (action_index == 0):
action['camera'] = [0, -2.5]
elif (action_index == 1):
action['camera'] = [0, 2.5]
elif (action_index == 2):
action['forward'] = 1
action['jump'] = 1
'''
if (action_index == 0):
action['camera'] = [0, -2.5]
action['attack'] = 1
elif (action_index == 1):
action['camera'] = [0, -2.5]
action['forward'] = 1
elif (action_index == 2):
action['camera'] = [0, -2.5]
action['left'] = 1
elif (action_index == 3):
action['camera'] = [0, -2.5]
action['right'] = 1
elif (action_index == 4):
action['camera'] = [0, -2.5]
action['back'] = 1
elif (action_index == 5):
action['camera'] = [0, -2.5]
action['jump'] = 1
elif (action_index == 6):
action['camera'] = [0, -2.5]
elif (action_index == 7):
action['camera'] = [0, 2.5]
action['attack'] = 1
elif (action_index == 8):
action['camera'] = [0, 2.5]
action['forward'] = 1
elif (action_index == 9):
action['camera'] = [0, 2.5]
action['left'] = 1
elif (action_index == 10):
action['camera'] = [0, 2.5]
action['right'] = 1
elif (action_index == 11):
action['camera'] = [0, 2.5]
action['back'] = 1
elif (action_index == 12):
action['camera'] = [0, 2.5]
action['jump'] = 1
elif (action_index == 13):
action['camera'] = [0, 2.5]
elif (action_index == 14):
action['camera'] = [-2.5, 0]
action['attack'] = 1
elif (action_index == 15):
action['camera'] = [-2.5, 0]
action['forward'] = 1
elif (action_index == 16):
action['camera'] = [-2.5, 0]
action['left'] = 1
elif (action_index == 17):
action['camera'] = [-2.5, 0]
action['right'] = 1
elif (action_index == 18):
action['camera'] = [-2.5, 0]
action['back'] = 1
elif (action_index == 19):
action['camera'] = [-2.5, 0]
action['jump'] = 1
elif (action_index == 20):
action['camera'] = [-2.5, 0]
elif (action_index == 21):
action['camera'] = [2.5, 0]
action['attack'] = 1
elif (action_index == 22):
action['camera'] = [2.5, 0]
action['forward'] = 1
elif (action_index == 23):
action['camera'] = [2.5, 0]
action['left'] = 1
elif (action_index == 24):
action['camera'] = [2.5, 0]
action['right'] = 1
elif (action_index == 25):
action['camera'] = [2.5, 0]
action['back'] = 1
elif (action_index == 26):
action['camera'] = [2.5, 0]
action['jump'] = 1
elif (action_index == 27):
action['camera'] = [2.5, 0]
elif (action_index == 28):
action['forward'] = 1
action['attack'] = 1
elif (action_index == 29):
action['forward'] = 1
action['jump'] = 1
elif (action_index == 30):
action['forward'] = 1
elif (action_index == 31):
action['jump'] = 1
elif (action_index == 32):
action['jump'] = 1
action['attack'] = 1
elif (action_index == 33):
action['back'] = 1
action['attack'] = 1
elif (action_index == 34):
action['back'] = 1
elif (action_index == 35):
action['left'] = 1
action['attack'] = 1
elif (action_index == 36):
action['left'] = 1
elif (action_index == 37):
action['right'] = 1
action['attack'] = 1
elif (action_index == 38):
action['right'] = 1
elif (action_index == 39):
action['sneak'] = 1
action['attack'] = 1
elif (action_index == 40):
action['sneak'] = 1
elif (action_index == 41):
action['attack'] = 1
#craft_action = nodes[node_index].crafting_agent.get_crafting_action()
#print("craft_action: ", craft_action)
#action = {**action, **craft_action}
#action['place'] = 'none'
#print("node_index: ", node_index)
#print("action: ", action)
#if node_index == 2:
# action['craft'] = 0
# if action['craft'] == 3:
# action['craft'] = 'none'
#
#if action['craft'] != 'none':
# print("action['craft']: ", action['craft'])
#print("")
observation_1, reward, done, info = env.step(action)
#reward += -0.005
inventory_channel_1 = np.zeros((64,64,1))
if 'inventory' in observation_1:
region_max_height = observation_1['pov'].shape[0]
region_max_width = observation_1['pov'].shape[1]
rs = 8
if min(region_max_height, region_max_width) < rs:
raise ValueError("'region_size' is too large.")
num_element_width = region_max_width // rs
inventory_channel_1 = np.zeros(shape=list(observation_1['pov'].shape[:-1]) + [1],
dtype=observation_1['pov'].dtype)
#print("state['inventory'].keys(): " + str(state['inventory'].keys()))
for key_idx, key in enumerate(observation_1['inventory'].keys()):
#print("key.shape : " + str(key))
#print("state['inventory'][key][i] : " + str(state['inventory'][key][i]))
item_scaled = np.clip(1 - 1 / (observation_1['inventory'][key] + 1), # Inversed
0, 1)
#print("item_scaled : " + str(item_scaled))
item_channel = np.ones(shape=[rs, rs, 1], dtype=observation_1['pov'].dtype) * item_scaled
width_low = (key_idx % num_element_width) * rs
height_low = (key_idx // num_element_width) * rs
if height_low + rs > region_max_height:
raise ValueError("Too many elements on 'inventory'. Please decrease 'region_size' of each component.")
inventory_channel_1[height_low:(height_low + rs), width_low:(width_low + rs), :] = item_channel
if nodes[node_index].name in item_list:
item_index = item_list.index(nodes[node_index].name)
else:
item_index = 0
scale = len(item_list)
item_layer = np.zeros([64,64,scale], dtype=np.float32)
item_layer[:, :, item_index] = 1.0
next_state = np.concatenate((observation_1['pov'] / 255.0, inventory_channel_1), axis=2)
next_state = np.concatenate((next_state, item_layer), axis=2)
next_state = tf.constant(next_state, dtype=tf.float32)
next_state = observation_1['pov'] / 255.0
'''
#print("node_index: ", node_index)
next_pov_array = observation_1['pov'] / 255.0
next_compassAngle_array = observation_1['compassAngle'] / 360.0
next_compassAngle_array = np.ones((64,64,1)) * next_compassAngle_array
next_state = np.concatenate((next_pov_array, next_compassAngle_array), 2)
next_state = np.expand_dims(next_state, 0)
'''
state = next_state
observation_ = observation_1
reward_sum += reward
step += 1
'''
if nodes[node_index].name not in item_list:
node_index += 1
elif observation_1['inventory'][nodes[node_index].name] != 0:
log_num = observation_1['inventory']['log']
planks_num = observation_1['inventory']['planks']
crafting_table_num = observation_1['inventory']['crafting_table']
stick_num = observation_1['inventory']['stick']
print("log_num: ", log_num)
print("planks_num: ", planks_num)
print("crafting_table_num: ", crafting_table_num)
print("stick_num: ", stick_num)
if node_index == 0:
if observation_1['inventory'][nodes[node_index].name] >= 4:
node_index += 1
else:
node_index += 1
print("node_index: ", node_index)
print("nodes[node_index].name: ", nodes[node_index].name)
print("")
'''
if done:
if arguments.env_id == 0:
scores.append(reward_sum)
episodes.append(episode_step)
average.append(sum(scores[-50:]) / len(scores[-50:]))
with writer.as_default():
tf.summary.scalar("average_reward", average[-1], step=episode_step)
writer.flush()
print("average_reward: " + str(average[-1]))
else:
print("reward_sum: " + str(reward_sum))
break
except (tf.errors.UnavailableError, tf.errors.CancelledError):
logging.info('Inference call failed. This is normal at the end of training.')