-
Notifications
You must be signed in to change notification settings - Fork 0
/
runner_dev.py
678 lines (564 loc) · 29.3 KB
/
runner_dev.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
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
# -*- coding: UTF-8 -*-#
# This runner is based on actions.FUNCTIONS (.\anaconda\a3_64\envs\py37_clone_v8\Lib\site-packages\pysc2\lib\actions.py)
from pysc2.lib import actions, features, units
from pysc2.env import sc2_env, run_loop, available_actions_printer
from pysc2 import maps
import sc2
from sc2 import run_game, maps, Race, Difficulty, position
from sc2.player import Bot, Computer, Human
from sc2.constants import NEXUS, PROBE, PYLON, ASSIMILATOR, GATEWAY, \
CYBERNETICSCORE, STALKER, STARGATE, VOIDRAY, OBSERVER, ROBOTICSFACILITY
# from terran_agent import TerranAgent
import models.models as models
import algorithms.q_learning as q_learning
from pathlib import Path
from absl import app, logging, flags
from matplotlib import pyplot as plt
import random
import math
import pickle
import os, sys
sys.path.append(os.path.dirname(os.path.dirname(os.path.realpath(__file__))))
import numpy as np
seed = 500
np.random.seed(seed)
_MOVE_RAND = 1000
_MOVE_MIDDLE = 2000
_BACKGROUND = 0
_AI_SELF = 1
_AI_ALLIES = 2
_AI_NEUTRAL = 3
_AI_HOSTILE = 4
_SELECT_ALL = [0]
_NOT_QUEUED = [0]
_QUEUED = [1]
EPS_START = 0.9
EPS_END = 0.025
EPS_DECAY = 2500
_PLAYER_RELATIVE = features.SCREEN_FEATURES.player_relative.index
_PLAYER_ID = features.SCREEN_FEATURES.player_id.index
_PLAYER_SELF = 1
_PLAYER_HOSTILE = 4
_AI_RELATIVE = features.SCREEN_FEATURES.player_relative.index
_AI_SELECTED = features.SCREEN_FEATURES.selected.index
_UNIT_TYPE = features.SCREEN_FEATURES.unit_type.index
_NO_OP = actions.FUNCTIONS.no_op.id
_MOVE = actions.FUNCTIONS.Scan_Move_screen.id
_MOVE_SCREEN = actions.FUNCTIONS.Attack_screen.id
# _MOVE_SCREEN = actions.FUNCTIONS.Attack_unit.id
_SELECT_ARMY = actions.FUNCTIONS.select_army.id
# _ATTACK_MINIMAP = actions.FUNCTIONS.Attack_Attack_minimap.id
_ATTACK_MINIMAP = actions.FUNCTIONS.Attack_minimap.id
army_selected = False
army_rallied = False
_SELECT_IDLE_WORKER = actions.FUNCTIONS.select_idle_worker.id
_TRAIN_SCV = actions.FUNCTIONS.Train_SCV_quick.id
_SELECT_POINT = actions.FUNCTIONS.select_point.id
_BUILD_SUPPLY_DEPOT = actions.FUNCTIONS.Build_SupplyDepot_screen.id
_BUILD_BARRACKS = actions.FUNCTIONS.Build_Barracks_screen.id
_TRAIN_MARINE = actions.FUNCTIONS.Train_Marine_quick.id
# _COLLECT_RESOURCES = actions.FUNCTIONS.Harvest_Gather_SCV_screen.id
_COLLECT_RESOURCES = actions.FUNCTIONS.Harvest_Gather_screen.id
_BUILD_MISSLE_TURRENT = actions.FUNCTIONS.Build_MissileTurret_screen.id
_BUILD_ENG_BAY = actions.FUNCTIONS.Build_EngineeringBay_screen.id
_TERRAN_COMMANDCENTER = 18
_TERRAN_SCV = 45
_TERRAN_MARINE = 48
_TERRAN_SUPPLY_DEPOT = 19
_TERRAN_BARRACKS = 21
_TERRAN_ENGINEERINGBAY = 22
_MissileTurret = 23
_NEUTRAL_BATTLESTATIONMINERALFIELD = 886,
_NEUTRAL_BATTLESTATIONMINERALFIELD750 = 887,
ACTION_DO_NOTHING = 'donothing'
ACTION_SELECT_SCV = 'selectscv'
ACTION_BUILD_SUPPLY_DEPOT = 'buildsupplydepot'
ACTION_BUILD_BARRACKS = 'buildbarracks'
ACTION_SELECT_BARRACKS = 'selectbarracks'
ACTION_TRAIN_MARINE = 'buildmarine'
ACTION_SELECT_ARMY = 'selectarmy'
ACTION_ATTACK = 'attack'
ACTION_COLLECT_RESOUCES = 'collect'
ACTION_BUILD_ENGBAY = 'buildengbay'
ACTION_BUILD_MISSLE_TURRENT = 'buildmissleturrent'
ACTION_TRAIN_SCV = 'trainscv'
categorical_actions = [
ACTION_SELECT_SCV,
ACTION_BUILD_SUPPLY_DEPOT,
ACTION_BUILD_BARRACKS,
ACTION_SELECT_BARRACKS,
ACTION_TRAIN_MARINE,
ACTION_BUILD_ENGBAY,
ACTION_BUILD_MISSLE_TURRENT,
ACTION_SELECT_ARMY,
ACTION_ATTACK,
ACTION_TRAIN_SCV,
ACTION_DO_NOTHING,
ACTION_COLLECT_RESOUCES
]
categorical_actions_id = [
_SELECT_IDLE_WORKER,
_BUILD_SUPPLY_DEPOT,
_BUILD_BARRACKS,
_SELECT_POINT,
_TRAIN_MARINE,
_SELECT_ARMY,
_NO_OP,
]
# spatial_actions = [ACTION_ATTACK]
spatial_actions = [_MOVE_SCREEN]
id_from_actions = {}
action_from_id = {}
# for ix, k in enumerate(spatial_actions):
# id_from_actions[k] = ix
# action_from_id[ix] = k
# for ix, k in enumerate(categorical_actions):
# id_from_actions[k] = ix+len(spatial_actions)
# action_from_id[ix+len(spatial_actions)] = k
for ix, k in enumerate(categorical_actions):
id_from_actions[k] = ix
action_from_id[ix] = k
FLAGS = flags.FLAGS
FLAGS(['run_sc2'])
# def get_action_v3(state):
#
# pass
def get_state(obs):
return [np.array(obs.observation['feature_screen']).reshape(1, 27, 64, 64),
np.array(obs.observation['feature_minimap']).reshape(1, 11, 64, 64),
np.array(obs.observation['player']).reshape(1, 11)
]
def to_yx(point):
"""transform a scalar from [0;4095] to a (y,x) coordinate in [0:63,0:63]"""
return int(point % 64), int((point - (point % 64)) / 64)
def transformLocation(obs, x, y):
player_y, player_x = (obs.observation['feature_minimap'][_PLAYER_RELATIVE] == _PLAYER_SELF).nonzero()
base_top_left = 1 if player_y.any() and player_y.mean() <= 31 else 0
if not base_top_left:
return [64 - x, 64 - y]
else:
return [x, y]
def get_action_v3(id_action, point, obs, num_dict=None):
# obs = obs[0]
unit_type = obs.observation['feature_screen'][_UNIT_TYPE]
depot_y, depot_x = (unit_type == _TERRAN_SUPPLY_DEPOT).nonzero()
supply_depot_exist = 1 if depot_y.any() else 0
if not supply_depot_exist:
num_dict['supply_deports'] = 0
barracks_y, barracks_x = (unit_type == _TERRAN_BARRACKS).nonzero()
barracks_exist = 1 if barracks_y.any() else 0
if not barracks_exist:
num_dict['barracks'] = 0
engbays_y, engbays_x = (unit_type == _TERRAN_ENGINEERINGBAY).nonzero()
engbays_exist = 1 if engbays_y.any() else 0
supply_limit = obs.observation['player'][4]
army_supply = obs.observation['player'][5]
food_workers = obs.observation['player'][6]
idle_workers_cnt = obs.observation['player'][7]
army_cnt = obs.observation['player'][8]
killed_unit_score = obs.observation['score_cumulative'][5]
killed_building_score = obs.observation['score_cumulative'][6]
current_state = np.zeros(20)
current_state[0] = supply_depot_exist
current_state[1] = barracks_exist
current_state[2] = supply_limit
current_state[3] = army_supply
hot_squares = np.zeros(16)
army_selected = False
army_rallied = False
if (obs.observation['feature_minimap'][_PLAYER_RELATIVE] == _PLAYER_HOSTILE).nonzero():
enemy_y, enemy_x = (obs.observation['feature_minimap'][_PLAYER_RELATIVE] == _PLAYER_HOSTILE).nonzero()
for i in range(0, len(enemy_y)):
y = int(math.ceil((enemy_y[i] + 1) / 16))
x = int(math.ceil((enemy_x[i] + 1) / 16))
hot_squares[((y - 1) * 4) + (x - 1)] = 1
for i in range(0, 16):
current_state[i + 4] = hot_squares[i]
smart_action = id_action
# if '_' in smart_action:
# smart_action, x, y = smart_action.split('_')
# (obs.observation['feature_minimap'][_PLAYER_RELATIVE] == _PLAYER_HOSTILE).nonzero()
if smart_action == ACTION_SELECT_SCV:
unit_type = obs.observation['feature_screen'][_UNIT_TYPE]
unit_y, unit_x = (unit_type == _TERRAN_SCV).nonzero()
if unit_y.any():
i = random.randint(0, len(unit_y) - 1)
target = [unit_x[i], unit_y[i]]
if _SELECT_IDLE_WORKER in obs.observation["available_actions"]:
func = actions.FunctionCall(_SELECT_IDLE_WORKER, [_NOT_QUEUED])
else:
func = actions.FunctionCall(_SELECT_POINT, [_NOT_QUEUED, target])
elif smart_action == ACTION_TRAIN_SCV:
# worker_cnt = num_dict["workers"]
# if _TRAIN_SCV in obs.observation['available_actions'] and worker_cnt < 16:
if _TRAIN_SCV in obs.observation['available_actions']:
func = actions.FunctionCall(_TRAIN_SCV, [_QUEUED])
# num_dict["workers"] += 1
elif smart_action == ACTION_COLLECT_RESOUCES:
# TODO: Warning about "必须以资源为目标"
unit_type = obs.observation['feature_screen'][_UNIT_TYPE]
scv_y, scv_x = (unit_type == units.Terran.SCV).nonzero()
mineral_y, mineral_x = (unit_type == units.Neutral.MineralField).nonzero()
# mineral_y, mineral_x = (unit_type == _NEUTRAL_BATTLESTATIONMINERALFIELD).nonzero()
if _COLLECT_RESOURCES in obs.observation['available_actions'] and idle_workers_cnt > 0:
if mineral_y.any() and scv_y.any():
i = random.randint(0, len(scv_y) - 1)
# target = (mineral_y[i], mineral_y[i])
# target = (mineral_y.mean(), mineral_y.mean())
# target = (scv_y.mean(), scv_x.mean())
target = (scv_y[i], scv_x[i])
# target = (11, 16)
func = actions.FunctionCall(_COLLECT_RESOURCES, [_NOT_QUEUED, target])
elif smart_action == ACTION_BUILD_SUPPLY_DEPOT:
if _BUILD_SUPPLY_DEPOT in obs.observation['available_actions']:
unit_type = obs.observation['feature_screen'][_UNIT_TYPE]
unit_y, unit_x = (unit_type == _TERRAN_COMMANDCENTER).nonzero()
if unit_y.any():
if supply_limit < 23:
target = (31, 8)
elif supply_limit < 31:
target = (26, 8)
elif supply_limit < 39:
target = (21, 8)
elif supply_limit < 47:
target = (16, 8)
# else:
# target = to_yx(point)
try:
func = actions.FunctionCall(_BUILD_SUPPLY_DEPOT, [_NOT_QUEUED, target])
return func, smart_action, num_dict
except UnboundLocalError:
# print(str(smart_action) + " " + str(point) + " is not an available action")
return get_action_v3(action_from_id[0], point, obs, num_dict) # 'selectscv'
elif smart_action == ACTION_BUILD_ENGBAY:
engbays_cnt = num_dict["engbays"]
if _BUILD_ENG_BAY in obs.observation['available_actions'] and not engbays_exist:
unit_type = obs.observation['feature_screen'][_UNIT_TYPE]
unit_y, unit_x = (unit_type == _TERRAN_COMMANDCENTER).nonzero()
if unit_y.any():
# target = to_yx(point)
target = (38, 44)
func = actions.FunctionCall(_BUILD_ENG_BAY, [_NOT_QUEUED, target])
try:
# num_dict["engbays"] += 1
return func, smart_action, num_dict
except UnboundLocalError:
# num_dict["engbays"] -= 1
# print(str(smart_action) + " " + str(point) + " is not an available action")
return get_action_v3(action_from_id[0], point, obs, num_dict) # 'selectscv'
elif smart_action == ACTION_BUILD_MISSLE_TURRENT:
missile_turrets_cnt = num_dict["missile_turrets"]
if _BUILD_MISSLE_TURRENT in obs.observation['available_actions'] and missile_turrets_cnt < 16:
unit_type = obs.observation['feature_screen'][_UNIT_TYPE]
unit_y, unit_x = (unit_type == _TERRAN_SUPPLY_DEPOT).nonzero()
if unit_y.any():
target = [(60, 16), (60, 26), (60, 36)]
if num_dict['missile_turrets'] == 0:
func = actions.FunctionCall(_BUILD_MISSLE_TURRENT, [_NOT_QUEUED, target[0]])
elif num_dict['missile_turrets'] == 1:
func = actions.FunctionCall(_BUILD_MISSLE_TURRENT, [_NOT_QUEUED, target[1]])
elif num_dict['missile_turrets'] == 2:
func = actions.FunctionCall(_BUILD_MISSLE_TURRENT, [_NOT_QUEUED, target[2]])
# else:
# target = to_yx(point)
# func = actions.FunctionCall(_BUILD_MISSLE_TURRENT, [_NOT_QUEUED, target])
try:
num_dict['missile_turrets'] += 1
return func, smart_action, num_dict
except UnboundLocalError:
num_dict['missile_turrets'] -= 1
# print(str(smart_action) + " " + str(point) + " is not an available action")
if engbays_exist:
return get_action_v3(action_from_id[0], point, obs, num_dict) # 'selectscv'
else:
return get_action_v3(action_from_id[5], point, obs,
num_dict) # 'buildengbay' # TODO: 无法建造导弹塔的原因不一定是因未建造工程港而未解锁，还有可能是前置动作未选择农民 # SOLVED #
elif smart_action == ACTION_BUILD_BARRACKS:
if _BUILD_BARRACKS in obs.observation['available_actions']:
unit_type = obs.observation['feature_screen'][_UNIT_TYPE]
unit_y, unit_x = (unit_type == _TERRAN_COMMANDCENTER).nonzero()
if unit_y.any() and num_dict["barracks"] < 3:
# target = to_yx(point)
if num_dict["barracks"] == 0:
target = (52, 18)
elif num_dict["barracks"] == 1:
target = (52, 28)
elif num_dict["barracks"] == 2:
target = (52, 38)
# else:
# target = to_yx(point)
try:
func = actions.FunctionCall(_BUILD_BARRACKS, [_NOT_QUEUED, target])
num_dict["barracks"] += 1
return func, smart_action, num_dict
except UnboundLocalError:
num_dict["barracks"] -= 1
# print(str(smart_action) + " " + str(point) + " is not an available action")
if num_dict['supply_deports'] == 0:
return get_action_v3(action_from_id[1], point, obs, num_dict) # 'buildsupplydepot'
else:
return get_action_v3(action_from_id[0], point, obs, num_dict) # 'selectscv'
elif smart_action == ACTION_SELECT_BARRACKS:
unit_type = obs.observation['feature_screen'][_UNIT_TYPE]
unit_y, unit_x = (unit_type == _TERRAN_BARRACKS).nonzero()
if unit_y.any():
# target = [int(unit_x.mean()), int(unit_y.mean())]
# target = (np.random.([(unit_x[i], unit_y[i]) for i in range(len(unit_x))]))
a_list = [(unit_x[i], unit_y[i]) for i in range(len(unit_x))]
target = list(map(lambda x: random.choice(a_list), range(1)))[0]
func = actions.FunctionCall(_SELECT_POINT, [_NOT_QUEUED, target])
try:
return func, smart_action, num_dict
except UnboundLocalError:
# print(str(smart_action) + " " + str(point) + " is not an available action")
return get_action_v3(action_from_id[2], point, obs, num_dict) # 'buildbarracks'
elif smart_action == ACTION_TRAIN_MARINE:
unit_type = obs.observation['feature_screen'][_UNIT_TYPE]
unit_y, unit_x = (unit_type == _TERRAN_BARRACKS).nonzero()
if _TRAIN_MARINE in obs.observation['available_actions'] and unit_y.any():
func = actions.FunctionCall(_TRAIN_MARINE, [_QUEUED])
try:
# num_dict["marines"] += 1
return func, smart_action, num_dict
except UnboundLocalError:
# num_dict["marines"] -= 1
# print(str(smart_action) + " " + str(point) + " is not an available action")
return get_action_v3(action_from_id[3], point, obs, num_dict) # 'selectbarracks'
elif smart_action == ACTION_SELECT_ARMY:
if _SELECT_ARMY in obs.observation['available_actions']:
func = actions.FunctionCall(_SELECT_ARMY, [_NOT_QUEUED])
try:
return func, smart_action, num_dict
except UnboundLocalError:
# print(str(smart_action) + " " + str(point) + " is not an available action")
return get_action_v3(action_from_id[4], point, obs, num_dict) # 'buildmarine'
elif smart_action == ACTION_ATTACK:
enemy_y, enemy_x = (obs.observation['feature_minimap'][_PLAYER_RELATIVE] == _PLAYER_HOSTILE).nonzero()
if (obs.observation['feature_minimap'][_PLAYER_RELATIVE] == _PLAYER_HOSTILE).nonzero(): # 攻击已知敌人
# for i in range(0, len(enemy_y)):
# marines_cnt = num_dict["marines"]
# if len(obs.observation['multi_select']) and army_cnt > 12 and num_dict['attack_cnt'] < 2:
if len(obs.observation['multi_select']) and army_cnt > 8:
# if obs.observation['multi_select'][0][0] != _TERRAN_SCV and _ATTACK_MINIMAP in obs.observation["available_actions"]:
if _ATTACK_MINIMAP in obs.observation["available_actions"]:
# if _ATTACK_MINIMAP in obs.observation["available_actions"]:
if enemy_y.any():
# target = [int(np.random.choice(enemy_x)), int(np.random.choice(enemy_y))]
target = to_yx(point) # TODO:
func = actions.FunctionCall(_ATTACK_MINIMAP, [_NOT_QUEUED, target])
# num_dict['marines'] = 0
# elif num_dict['attack_cnt'] >= 2 and len(obs.observation['multi_select']) and army_cnt >= 3:
# # if obs.observation['multi_select'][0][0] != _TERRAN_SCV and _ATTACK_MINIMAP in obs.observation["available_actions"]:
# if _ATTACK_MINIMAP in obs.observation["available_actions"]:
# # if _ATTACK_MINIMAP in obs.observation["available_actions"]:
# if enemy_y.any():
# target = [int(np.random.choice(enemy_x)), int(np.random.choice(enemy_y))]
# # target = to_yx(point) # TODO:
# func = actions.FunctionCall(_ATTACK_MINIMAP, [_NOT_QUEUED, target])
# # num_dict['marines'] = 0
# else:
# if len(obs.observation['multi_select']):
# # if obs.observation['multi_select'][0][0] != _TERRAN_SCV and _ATTACK_MINIMAP in obs.observation["available_actions"]:
# if _ATTACK_MINIMAP in obs.observation["available_actions"]:
# # if _ATTACK_MINIMAP in obs.observation["available_actions"]:
# if enemy_y.any():
# target = [int(np.random.choice(enemy_x)), int(np.random.choice(enemy_y))]
# # target = to_yx(point) # TODO:
# func = actions.FunctionCall(_ATTACK_MINIMAP, [_NOT_QUEUED, target])
# # num_dict['marines'] = 0
# num_dict['attack_cnt'] += 1
else: # 攻击任意位置（未找到敌人时，类似巡逻）
if len(obs.observation['multi_select']):
# if obs.observation['multi_select'][0][0] != _TERRAN_SCV and _ATTACK_MINIMAP in obs.observation["available_actions"]:
if _ATTACK_MINIMAP in obs.observation["available_actions"]:
target = to_yx(point)
func = actions.FunctionCall(_ATTACK_MINIMAP, [_NOT_QUEUED, target])
try:
num_dict['attack_cnt'] += 1
return func, smart_action, num_dict
except UnboundLocalError:
num_dict['attack_cnt'] -= 1
# print(str(smart_action) + " " + str(point) + " is not an available action")
if army_cnt < 8:
return get_action_v3(action_from_id[4], point, obs, num_dict) # 'buildmarine'
else:
return get_action_v3(action_from_id[7], point, obs, num_dict) # 'select_army'
elif smart_action == ACTION_DO_NOTHING:
func = actions.FunctionCall(_NO_OP, [])
try:
return func, smart_action, num_dict
except UnboundLocalError:
# print(str(smart_action) + " " + str(point) + " is not an available action")
return actions.FunctionCall(_NO_OP, []), ACTION_DO_NOTHING, num_dict
def main(unused_argv):
viz = True
replay_prefix = 'D:/software/python_prj/SCII/SCII_Bots/replays/deterministic_sequence'
replay_dir = '/replays'
real_time = False
ensure_available_actions = True
disable_fog = True
train_mode = True # True False
game_steps_per_episode = 5000 # 0 actually means unlimited
if train_mode == True:
MAX_EPISODES = 1000
else:
MAX_EPISODES = 1
MAX_STEPS = 3000 # 运行500个step耗时约为2：57
try:
# run trajectories and train
with sc2_env.SC2Env(
map_name="Simple64",
players=[sc2_env.Agent(sc2_env.Race.terran),
# sc2_env.Bot(sc2_env.Race.protoss, sc2_env.Difficulty.easy)
sc2_env.Bot(sc2_env.Race.protoss, sc2_env.Difficulty.very_easy)
],
visualize=viz, agent_interface_format=sc2_env.AgentInterfaceFormat(
use_raw_units=True,
feature_dimensions=sc2_env.Dimensions(
screen=64,
minimap=64)),
random_seed=seed,
# discount=0.3,
realtime=real_time,
ensure_available_actions=ensure_available_actions,
disable_fog=disable_fog,
# save_replay_episodes=1,
# replay_prefix=replay_prefix,
# replay_dir=replay_dir,
# game_steps_per_episode=game_steps_per_episode
) as env:
done = False
# history = []
dataset = []
score = 0
score_pre = 0
max_batch_pool_in_last_play = 0
max_episode_in_last_play = 0
path_lst = os.listdir('./save/dqn')
if len(path_lst) != 0:
max_episode_in_last_play = max([int(p.split('.')[0].split('i')[-1]) for p in path_lst])
load_path = Path(Path(os.getcwd()) / 'save' / 'dqn' / 'Simple64-dqn-epi{}.pt'.format(max_episode_in_last_play))
else:
load_path = 'none'
algo = q_learning.DeepQLearning(load_path)
logs_path_lst = os.listdir('./logs')
if len(logs_path_lst) != 0:
max_batch_pool_in_last_play = max([int(os.listdir('./logs')[p].split('p')[-2].split('.')[0]) for p in range(len(logs_path_lst))])
batch_pool_idx = max_batch_pool_in_last_play + 1
losses_lst = []
for e in range(max_episode_in_last_play+1, max_episode_in_last_play+MAX_EPISODES+1):
if e > 0:
obs = env.reset()
score = 0
score_pre = 0
control_seq = []
control_idx = 1
num_dict = {"workers": 0, "idle_workers": 0, "barracks": 0, "engbays": 0,
# "marines": 0,
"missile_turrets": 0, 'attack_cnt': 0}
for time in range(MAX_STEPS):
init = False
if e == 0 and time == 0:
init = True
state_model = [np.array(obs[0].observation.feature_screen),
np.array(obs[0].observation.feature_minimap), np.array(obs[0].observation.player)]
# TODO: state_model = [np.array(obs[0].observation.feature_screen), np.array(obs[0].observation.feature_minimap), np.array(obs[0].observation.player), np.array(obs[0].observation.last_actions)]
# action, point = action_from_id[np.random.choice(len(action_from_id), 1)[0]], np.random.randint(4096)
# func, actual_action, new_num_dict = get_action_v3(action, point, obs=obs[0], num_dict=num_dict)
preds = algo.choose_action_v(state_model, init)
# print()
action, point = action_from_id[np.argmax(preds[0].detach().cpu().numpy())], \
[i for i in range(4096)][np.argmax(preds[1].detach().cpu().numpy())]
func, actual_action, new_num_dict = get_action_v3(action, point, obs=obs[0], num_dict=num_dict)
next_obs = env.step([func])
# print(actual_action, point)
next_state = get_state(next_obs[0])
num_dict = new_num_dict
state_model_next = [np.array(obs[0].observation.feature_screen),
np.array(obs[0].observation.feature_minimap),
np.array(obs[0].observation.player)]
# reward_a = float(next_obs[0].reward) + float(next_obs[0].observation.score_cumulative[
# 0]) * 10e-8 # next_obs[0].observation.score_cumulative[0]: 'score' (2745598944344)
# reward_a = float(next_obs[0].observation.score_cumulative[11]) * 10e-8 # spent_minerals
# reward_p = float(
# next_obs[0].observation.score_cumulative[5] + next_obs[0].observation.score_cumulative[6]) * 10e-8 # next_obs[0].observation.score_cumulative[5], [6]: 'killed_value_units' (2745642291968)，'killed_value_structures' (2745642292040)
reward_a = float(next_obs[0].observation.score_cumulative[11]) * 10e-2 # spent_minerals
reward_p = float(
next_obs[0].observation.score_cumulative[5] + next_obs[0].observation.score_cumulative[6]) # next_obs[0].observation.score_cumulative[5], [6]: 'killed_value_units' (2745642291968)，'killed_value_structures' (2745642292040)
if actual_action == action:
reward_a = reward_a + 5
reward = [reward_a, reward_p]
last_action = obs[0].observation.last_actions
if env._controllers and env._controllers[0].status.value != 3:
done = True
if env._obs[0].player_result[0].result == 1: # player0(unknown)胜利
reward = list(np.array(reward) + 10000)
elif env._obs[0].player_result[0].result == 2: # player0(unknown)战败
reward = list(np.array(reward) - 10000)
# dataset = []
if time == MAX_STEPS - 1:
done = True
score += reward_a
if done:
reward = list(np.array(reward) - 5000)
dataset.append(
[e, time, state_model, state_model_next, action, actual_action, last_action, point, reward,
score, done]
)
num_dict["barracks"] = 0
print("episode: {}/{}, score: {}".format(e, max_episode_in_last_play+MAX_EPISODES, score))
if score_pre < score:
score_pre = score
# history_arr = np.array(history)
# np.savez_compressed('./logs/history_dqn_sequence_{}.npz'.format(str(e)), history_arr)
# history = []
done = False
# logging.info("state: %s", str(state))
# logging.info("action: %s", str(action))
# logging.info("reward: %s", str(obs.reward))
previous_state = state
previous_action = action
algo.copy()
break
# # save agent info
# history.append(
# [e, time, state_model, state_model_next, action, actual_action, last_action, point, reward, score, done]
# )
# if len(history) == 128:
# minibatch = random.sample(history, 32)
dataset.append(
[e, time, state_model, state_model_next, action, actual_action, last_action, point, reward,
score, done]
)
if len(dataset) >= 256: # TODO: HOW TO LEARN ??
if train_mode:
loss_per_batch = algo.learn(dataset, id_from_actions)
losses_lst.append(np.mean(loss_per_batch))
if e % 10 == 0:
# np.savez_compressed('./logs/history_dqn_sequence_bp{}.npz'.format(str(batch_pool_idx)), np.array(dataset))
# 保存为pickle文件
with open('./logs/history_dqn_sequence_bp{}_score{}.pkl'.format(str(batch_pool_idx), str(score)), "wb") as f:
pickle.dump(np.array(dataset), f)
dataset = []
batch_pool_idx += 1
# algo.check_state_exist(state_model)
state = next_state
obs = next_obs
# time += 1
if train_mode:
save_path = './save/dqn/Simple64-dqn-epi{}-score{}.pt'.format(e, score)
algo.save(save_path)
plt.plot(losses_lst)
plt.show()
except KeyboardInterrupt:
pass
# finally:
# elapsed_time = time.time() - start_time
# print("Took %.3f seconds for %s steps: %.3f fps" % (
# elapsed_time, total_frames, total_frames / elapsed_time))
if __name__ == '__main__':
app.run(main)
"""
I0606 19:39:39.231369 78480 sc2_env.py:725] Episode 54 finished after 12248 game steps. Outcome: [1], reward: [1], score: [6465]
episode: 54/1000, score: 192297.5
"""