-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathPPO_HardAtariGame.py
201 lines (132 loc) · 6.81 KB
/
PPO_HardAtariGame.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
import os
import gym
import torch
from agents import TYPE
from agents.PPOAtariAgent import PPOAtariAgent, PPOAtariRNDAgent, PPOAtariForwardModelAgent, PPOAtariQRNDAgent, \
PPOAtariDOPAgent, PPOAtariSRRNDAgent, PPOAtariDOPAgent, PPOAtariCNDAgent, PPOAtariFWDAgent, PPOAtariICMAgent, PPOAtariFEDRefAgent
from experiment.ppo_nenv_experiment import ExperimentNEnvPPO
from plots.paths import models_root
from utils.AtariWrapper import WrapperHardAtari
from utils.MultiEnvWrapper import MultiEnvParallel
def encode_state(state):
return torch.tensor(state, dtype=torch.float32)
def test(config, path, env_name):
env = WrapperHardAtari(gym.make(env_name, render_mode='human'))
input_shape = env.observation_space.shape
action_dim = env.action_space.n
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariCNDAgent(input_shape, action_dim, config, TYPE.discrete)
agent.load(os.path.join(models_root, path))
experiment.test(agent)
env.close()
def run_baseline(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_baseline(agent, trial)
env.close()
def run_rnd_model(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariRNDAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_rnd_model(agent, trial)
env.close()
def run_qrnd_model(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariQRNDAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_qrnd_model(agent, trial)
env.close()
def run_sr_rnd_model(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariSRRNDAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_sr_rnd_model(agent, trial)
env.close()
def run_cnd_model(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariCNDAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_cnd_model(agent, trial)
env.close()
def run_dop_model(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariDOPAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_dop_model(agent, trial)
env.close()
def run_fed_ref_model(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariFEDRefAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_fed_ref_model(agent, trial)
env.close()
def run_forward_model(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariForwardModelAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_forward_model(agent, trial)
env.close()
def run_fwd_model(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariFWDAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_fwd_model(agent, trial)
env.close()
def run_icm_model(config, trial, env_name):
print('Creating {0:d} environments'.format(config.n_env))
env = MultiEnvParallel([WrapperHardAtari(gym.make(env_name)) for _ in range(config.n_env)], config.n_env, config.num_threads)
input_shape = env.observation_space.shape
action_dim = env.action_space.n
print('Start training')
experiment = ExperimentNEnvPPO(env_name, env, config)
experiment.add_preprocess(encode_state)
agent = PPOAtariICMAgent(input_shape, action_dim, config, TYPE.discrete)
experiment.run_icm_model(agent, trial)
env.close()