-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathppo2_contra_baselines_agent (lvl 2 easy clear).py
210 lines (173 loc) · 8.81 KB
/
ppo2_contra_baselines_agent (lvl 2 easy clear).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
#!/usr/bin/env python
"""
Train an agent on contra III using PPO implementation by OpenAI baselines
"""
try:
from mpi4py import MPI
except ImportError:
MPI = None
import tensorflow as tf
import baselines.ppo2.ppo2 as ppo2
from baselines.common.atari_wrappers import *
from baselines.common.retro_wrappers import *
from baselines import bench
from baselines.common.vec_env.subproc_vec_env import SubprocVecEnv
from baselines.common.vec_env.dummy_vec_env import DummyVecEnv
from baselines.common import set_global_seeds
import time
import retro
import gym
from baselines.common.models import register
from baselines.a2c import utils
from baselines.a2c.utils import conv, fc, conv_to_fc, batch_to_seq, seq_to_batch
import numpy as np
import argparse
import os
from baselines import logger
os.environ['OPENAI_LOG_FORMAT']='stdout,log,csv,tensorboard'
def main():
"""Run PPO until the environment throws an exception."""
config = tf.ConfigProto()
config.gpu_options.allow_growth = True # pylint: disable=E1101
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--num_env',default=1,type=int)
parser.add_argument('--seed', default=None,type=int)
parser.add_argument('--game',default='ContraIII-Snes')
parser.add_argument('--state', default='level1.1player.easy.100lives') #state=retro.State.DEFAULT
parser.add_argument('--scenario', default='scenario')
parser.add_argument('--discrete_actions', default=0, type=int)
parser.add_argument('--bk2dir', default='videos')
parser.add_argument('--monitordir', default='logs')
parser.add_argument('--sonic_discretizer', default=1, type=int)
parser.add_argument('--clip_rewards', default=0, type=int)
parser.add_argument('--stack', default=4,type=int)
parser.add_argument('--time_limit', default=8000,type=int)
parser.add_argument('--scale_reward', default=0.01,type=float)
parser.add_argument('--warp_frame', default=1,type=int)
parser.add_argument('--stochastic_frame_skip', default=4,type=int)
parser.add_argument('--skip_prob', default=0.0,type=float)
parser.add_argument('--network', default='cnn')
parser.add_argument('--scenario_number', default=1,type=int)
parser.add_argument('--load_path',default=None)
args = parser.parse_args()
time_int = int(time.time())
env_vec = make_vec_env(args,time_int)
logger.configure(dir='./log/{}'.format(time_int), format_strs=['stdout', 'log', 'csv', 'tensorboard'])
with tf.Session(config=config):
ppo2.learn(network=args.network, #network='contra_net', #network='cnn',
env=env_vec,
nsteps=256,#1024,
nminibatches=32,#16,256,512,64,128
lam=0.95,
gamma=0.997, #0.99
noptepochs=3, #3,
log_interval=100,
ent_coef=0.003,#0.003,#0.003, 0.001, 0.005 #many actions #0.01
lr=lambda _: 5e-4,#2e-4,1e-4,5e-5
cliprange=0.20,
save_interval=100,
seed=args.seed,
vf_coef=0.5,
max_grad_norm=0.5,
save_path='ppo_save/{}'.format(time_int),
#load_path=args.load_path,
total_timesteps=int(2e8))
#total_timesteps=int(10000))
def make_vec_env(args, time_int, start_index=0):
"""
Create a wrapped, monitored SubprocVecEnv
"""
mpi_rank = MPI.COMM_WORLD.Get_rank() if MPI else 0
seed = args.seed + 10000 * mpi_rank if args.seed is not None else None
def make_thunk(rank):
return lambda: make_env(time_int=time_int,seed=args.seed,subrank=rank, game=args.game,state=args.state,scenario=args.scenario,
discrete_actions=bool(args.discrete_actions),bk2dir=args.bk2dir,monitordir=args.monitordir,
sonic_discretizer=bool(args.sonic_discretizer),clip_rewards=bool(args.clip_rewards), stack=args.stack,
time_limit=args.time_limit,scale_reward=args.scale_reward,warp_frame=bool(args.warp_frame),
stochastic_frame_skip=args.stochastic_frame_skip,skip_prob=args.skip_prob,scenario_number=args.scenario_number)
set_global_seeds(seed)
if args.num_env > 1:
return SubprocVecEnv([make_thunk(i + start_index) for i in range(args.num_env)])
else:
return DummyVecEnv([make_thunk(start_index)])
def make_env(time_int, seed=None, subrank=0, game='ContraIII-Snes', state='level1.1player.easy.100lives', scenario='scenario', discrete_actions=False,
bk2dir='videos',monitordir='logs', sonic_discretizer=True, clip_rewards=False,
stack=4,time_limit=8000,scale_reward=0.01,warp_frame=True,stochastic_frame_skip=4,skip_prob=0.0,scenario_number=1):
use_restricted_actions = retro.Actions.FILTERED # retro.ACTIONS_FILTERED
if discrete_actions:
use_restricted_actions = retro.Actions.DISCRETE # retro.ACTIONS_DISCRETE
if scenario_number <= 1:
print("scenario is {}".format(scenario))
env = retro.make(game, state, scenario=scenario, use_restricted_actions=use_restricted_actions)
else:
scenario = scenario + "_{}".format(subrank%scenario_number)
print("scenario is {}".format(scenario))
env = retro.make(game, state, scenario=scenario, use_restricted_actions=use_restricted_actions)
if bk2dir:
#env.auto_record(bk2dir)
bk2dir = bk2dir + "/{}_{}".format(time_int,subrank)
print(bk2dir)
if not os.path.exists(bk2dir):
os.makedirs(bk2dir)
env = MovieRecord(env,bk2dir,k=10)#k is how often to save and record the video
if monitordir:
#print("test path {}".format(os.path.join(monitordir, 'log_{}_{}'.format(time_int,subrank))))
env = bench.Monitor(env,os.path.join(monitordir, 'log_{}_{}'.format(time_int,subrank)),allow_early_resets=True)
if stochastic_frame_skip:
#print("skip prob is {}".format(skip_prob))
env = StochasticFrameSkip(env,stochastic_frame_skip,skip_prob)
if time_limit:
env = TimeLimit(env,max_episode_steps=time_limit)
if sonic_discretizer:
#env = SonicDiscretizer(env)
env = ContraDiscretizer(env)
if scale_reward:
env = RewardScaler(env,scale=scale_reward)
if clip_rewards:
env = ClipRewardEnv(env)
if warp_frame:
env = WarpFrame(env,width=112,height=128) #,grayscale=False
if stack:
env = FrameStack(env, stack)
env.seed(seed + subrank if seed is not None else None)
return env
class ContraDiscretizer(gym.ActionWrapper):
"""
Wrap a gym-retro environment and make it use discrete
actions for the Sonic game.
"""
def __init__(self, env):
super(ContraDiscretizer, self).__init__(env)
# SNES keys
buttons = ["B", "Y", "SELECT", "START", "UP", "DOWN", "LEFT", "RIGHT", "A", "X", "L", "R"]
#lvl 2
actions = [['Y'], ['UP'], ['RIGHT'], ['DOWN'], ['LEFT'], ['A'], ['R'], ['L'],
# 3 button combos optional but might be useful for harder parts
['Y','R','UP'],['Y','R','RIGHT'],['Y','R','DOWN'],['Y','R','LEFT'],
['Y', 'L', 'UP'], ['Y', 'L', 'RIGHT'], ['Y', 'L', 'DOWN'], ['Y', 'L', 'LEFT'],
['Y','X'], ['Y','UP'], ['Y','RIGHT'], ['Y','DOWN'], ['Y','LEFT'], ['Y','R'], ['Y','L'], ['Y','B']]
#level 3
# actions = [['Y'], ['UP'], ['RIGHT'], ['DOWN'], ['LEFT'], ['A'],
# #['Y', 'DOWN', 'B'],
# ['Y', 'LEFT'], ['Y', 'RIGHT'], ['Y', 'X'], ['Y', 'UP'], ['Y', 'DOWN'],
# ['Y', 'B', 'LEFT'], ['Y', 'B', 'RIGHT'], ['Y', 'UP', 'RIGHT'], ['Y', 'DOWN', 'RIGHT'],
# ['Y', 'UP', 'LEFT'], ['Y', 'DOWN', 'LEFT']
# ] #
#['Y', 'R', 'UP', 'LEFT'], ['Y', 'R', 'UP', 'RIGHT'], ['Y', 'R', 'DOWN', 'LEFT'], ['Y', 'R', 'DOWN', 'RIGHT']
#['Y', 'L', 'R', 'RIGHT'], ['Y', 'L', 'R', 'LEFT'], ['Y', 'UP', 'LEFT'], ['Y', 'DOWN', 'LEFT'], ['A'],
# ['Y', 'R', 'UP', 'LEFT'], ['Y', 'R', 'UP', 'RIGHT'], ['Y', 'X', 'R', 'UP', 'LEFT'], ['Y', 'X', 'R', 'UP', 'RIGHT'], #level 4 specific
#23 actions, more than I would like but certain ones needed at specific moments
self._actions = []
for action in actions:
arr = np.array([False] * 12)
if action == ['NOOP']:
self._actions.append(arr)
continue
for button in action:
arr[buttons.index(button)] = True
self._actions.append(arr)
self.action_space = gym.spaces.Discrete(len(self._actions))
def action(self, a): # pylint: disable=W0221
return self._actions[a].copy()
if __name__ == '__main__':
main()