-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
215 lines (195 loc) · 6.75 KB
/
train.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
from rerun import get_last_stamp, get_stamps, get_starting_episode
from model import make_model, make_baseline_model
from malmo_agent import MalmoAgent
import malmo.MalmoPython as MalmoPython
from pathlib import Path
import json
import sys
import time
import numpy as np
# Constants
xml_file = "./envs/zombie_fight.xml"
episodes = 5000
baseline = True
video_shape = (480, 640, 3)
input_shape = (84, 112, 3)
save = True
load_last_trained = True
load_model = None
start_episode = 1
max_steps_per_episode = 1000
running_average_length = episodes // 20
num_zombies = 2
agent_cfg = {
"alpha": 0.0005,
"gamma": 0.85,
"batch_size": 64,
"epsilon": 1,
"epsilon_decay": 0.992,
"epsilon_min": 0.05,
"copy_period": 300,
"mem_size": 5000
}
# agent_cfg = {
# "alpha": 0.0005,
# "gamma": 0.85,
# "batch_size": 64,
# "epsilon": 1,
# "epsilon_decay": 0.998,
# "epsilon_min": 0.05,
# "copy_period": 300,
# "mem_size": 10000
# }
# agent_cfg = {
# "alpha": 0.0005,
# "gamma": 0.92,
# "batch_size": 64,
# "epsilon": 1,
# "epsilon_decay": 0.998,
# "epsilon_min": 0.05,
# "copy_period": 100,
# "mem_size": 10000
# }
if __name__ == "__main__":
# Runtime Generated Constants
stamp = int(time.time()) if not load_last_trained else get_last_stamp()
if load_model is not None:
stamp = get_stamps()[load_model]
env_name = xml_file.split("/")[-1].split(".")[0]
if baseline:
env_name += "_baseline"
env_name += "_" + str(num_zombies)
model_file = f"models/{env_name}_{stamp}.h5"
metric_file = f"metrics/{env_name}_{stamp}.json"
h, w, d = video_shape
input_shape = (224, 224, 3) if baseline else input_shape
n = len(MalmoAgent.actions)
r = lambda x: np.around(x, decimals=3)
# Environment Setup
agent_host = MalmoPython.AgentHost()
try:
agent_host.parse(sys.argv)
except RuntimeError as e:
print("ERROR:", e)
print(agent_host.getUsage())
exit(1)
if agent_host.receivedArgument("help"):
print(agent_host.getUsage())
exit(0)
agent_host.setObservationsPolicy(
MalmoPython.ObservationsPolicy.LATEST_OBSERVATION_ONLY
)
agent_host.setVideoPolicy(MalmoPython.VideoPolicy.LATEST_FRAME_ONLY)
xml = Path(xml_file).read_text()
xml = xml.replace("{{width}}", str(w)).replace("{{height}}", str(h))
mission = MalmoPython.MissionSpec(xml, True)
record = MalmoPython.MissionRecordSpec()
# Agent Setup
model = (
make_baseline_model(video_shape, n) if baseline else make_model(input_shape, n)
)
print(model.summary())
agent = MalmoAgent(
alpha=agent_cfg["alpha"],
gamma=agent_cfg["gamma"],
batch_size=agent_cfg["batch_size"],
epsilon=agent_cfg["epsilon"],
epsilon_decay=agent_cfg["epsilon_decay"],
epsilon_min=agent_cfg["epsilon_min"],
copy_period=agent_cfg["copy_period"],
mem_size=agent_cfg["mem_size"],
model=model,
model_file=model_file,
metric_file=metric_file,
input_shape=input_shape,
agent_host=agent_host,
)
if load_last_trained or load_model is not None:
agent.load_model()
agent.load_data()
start_episode = get_starting_episode(agent)
agent.epsilon = agent.epsilon_decay ** (start_episode - 1)
elif load_model:
agent.load_model(load_model)
# Episode Loop
for ep in range(start_episode - 1, episodes):
# Mission Setup
first = True
working = False
while not working:
if not first:
print("Had to restart the mission since the data was not sent to the server!")
time.sleep(30)
first = False
max_retries = 3
for retry in range(max_retries):
try:
agent_host.startMission(mission, record)
break
except RuntimeError as e:
if retry == max_retries - 1:
print("Error starting mission:", e)
exit(1)
else:
time.sleep(2)
world_state = agent_host.getWorldState()
i = 0
while not world_state.has_mission_begun and i < 500:
time.sleep(0.1)
world_state = agent_host.getWorldState()
for error in world_state.errors:
print("Error:", error.text)
i += 1
working = i < 500
for _ in range(num_zombies):
agent_host.sendCommand(
"chat /summon Zombie "
+ str(np.random.randint(-10, 11))
+ " 4.5 "
+ str(np.random.randint(-10, 11))
+ " {HealF:10.0f}"
)
agent_host.sendCommand("chat /gamerule naturalRegeneration false")
agent_host.sendCommand("chat /difficulty 1")
# Step Loop
step = 0
start_time = time.time()
state = None
action = 0
reward = 0
done = False
while world_state.is_mission_running and step < max_steps_per_episode:
agent_host.sendCommand("attack 1")
time.sleep(0.02)
world_state = agent_host.getWorldState()
if len(world_state.observations) and len(world_state.video_frames):
obs = json.loads(world_state.observations[-1].text)
frame = world_state.video_frames[0]
next_state = agent.process_frame(frame)
reward, done = agent.process_observation(obs)
if state is not None:
agent.remember(state, action, reward, next_state, done)
agent.learn()
state = next_state
action = agent.choose_and_take_action(state)
step += 1
time_elapsed = time.time() - start_time
print(
f"Step {1 + step}; Reward {r(reward)}; Score {r(agent.temp['cumulative_reward'])}; Kills {r(agent.temp['kills'])}; Epsilon {r(agent.epsilon)}; Time Elapsed {int(time_elapsed)}s"
+ " " * 20,
end="\r",
)
agent.finished_episode()
agent.metrics["times"].append(time_elapsed)
avg_score = np.mean(
agent.metrics["cumulative_rewards"][
max(0, ep - running_average_length) : ep + 1
]
)
print(
f"Episode {ep + 1} of {episodes}; Score {r(agent.metrics['cumulative_rewards'][-1])}; Kills {r(agent.metrics['kills'][-1])}; Average Score {r(avg_score)}; Episode Time {int(time_elapsed)}s"
+ " " * 20
)
if save:
agent.save_model()
agent.save_data()