-
Notifications
You must be signed in to change notification settings - Fork 0
/
loc8.py
146 lines (97 loc) · 3.22 KB
/
loc8.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
"""Loc8."""
import matplotlib.pyplot as plt
import numpy as np
import rl.core as krl
from model import model
LAYERS = 3
class Loc8World:
"""Loc8 world simulator."""
def __init__(self, *shape, goal=None, start=None):
assert shape, "World must have a defined shape"
self.shape = np.array(shape)
self.position = np.array(start) if start else np.zeros(len(shape))
self.array = np.zeros((*self.shape, LAYERS))
self._goal = goal
self.goal = self._goal or self.random_goal()
def move(self, delta):
copy = self.position.copy()
self.position += delta
self.position = self.position.clip(0, self.shape - 1)
return copy
def coords(self, position=None):
if position is None:
position = self.position
return tuple(position.astype(np.int))
@property
def distance_to_goal(self):
"""Return the current distance to the goal."""
return np.linalg.norm(self.goal - self.position)
def random_goal(self):
"""Choose a random goal."""
return np.array([np.random.randint(axis) for axis in self.shape])
class Loc8Env(krl.Env):
"""Loc8 environment."""
def __init__(self, *shape, vision=2, goal=None):
self.world = Loc8World(*shape, goal=goal)
self.vision = vision
self.choices = np.array(
np.meshgrid(*[[-1, 0, 1]] * len(shape))
).T.reshape(-1, len(shape))
def observe(self):
"""Retrieve observations for reinforcement learning."""
coords = self.world.coords()
self.world.array[(*coords, 1)] = 1
world = self.world.array.copy()
world[(*coords, 2)] = 1
return world
def step(self, action):
reward = 0
choice = self.choices[action]
self.world.move(choice)
observation = self.observe()
done = self.world.distance_to_goal < self.vision
reward = 100 if done else 0
# observation, reward, done, info
return observation, reward, done, {}
def reset(self):
plt.clf()
plt.xlim(0, self.world.shape[0])
plt.ylim(0, self.world.shape[1])
self.world = Loc8World(*self.world.shape, goal=self.world._goal)
goal = self.world.goal
plt.plot([goal[0]], [goal[1]], 'g^', markersize=20)
return self.observe()
########
def render(self, mode="human", close=False):
if mode != "human":
return
plt.plot(*np.expand_dims(self.world.position, 0).T, 'o')
plt.pause(0.01)
########
def close(self):
pass
###
def run(*shape, dense_layers=16, **kwargs):
"""Run reinforcement learning algorithm with a given world size."""
env = Loc8Env(*shape, **kwargs)
nb_actions = len(env.choices)
observation_shape = (*env.world.shape, LAYERS)
agent = model(
nb_actions, observation_shape,
dense_layers=dense_layers
)
agent.fit(
env,
nb_steps=50000,
nb_max_episode_steps=100,
visualize=True,
verbose=1,
)
agent.test(
env,
nb_episodes=50,
nb_max_episode_steps=100,
visualize=True,
)
if __name__ == "__main__":
run(10, 10, goal=(2, 8))