-
Notifications
You must be signed in to change notification settings - Fork 0
/
cartpole.py
141 lines (105 loc) · 4.38 KB
/
cartpole.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
import numpy as np
from collections import defaultdict
import math
class CartPole(object):
"""
This Class defines the defaul behavior of CartPole agent
"""
def __init__(self, action_space):
self.action_space = action_space
def act(self, **kwargs):
""" Returns random action from action space"""
return self.action_space.sample()
def learn(self, **kwargs):
raise NotImplementedError
class RandomSearch(CartPole):
def __init__(self, action_space, reward_tresh=50, parameters=None):
super().__init__(action_space)
if parameters is None:
self.reward_tresh = reward_tresh
self.parameters = np.random.random(4) * 2 - 1
self.best_parameters = []
else:
assert len(
parameters) == 4, "parameters should be an array of length 4"
self.parameters = parameters
def act(self, state):
assert len(state) == 4, "state should be an array of leng 4"
action = np.dot(self.parameters, state)
if action > 0:
return 1
else:
return 0
def learn(self, reward, done):
if reward >= self.reward_tresh:
self.reward_tresh = reward
self.best_parameters.append((self.parameters, reward))
if done:
self.parameters = np.random.random(4) * 2 - 1
class QlearningAgent(CartPole):
## Learning related constants
MIN_EXPLORE_RATE = 0.01
MIN_LEARNING_RATE = 0.1
def get_explore_rate(self, t, MIN_EXPLORE_RATE=MIN_EXPLORE_RATE):
return max(MIN_EXPLORE_RATE, min(1, 1.0 - math.log10((t + 1) / 25)))
def get_learning_rate(self, t, MIN_LEARNING_RATE=MIN_LEARNING_RATE):
return max(MIN_LEARNING_RATE, min(0.5, 1.0 - math.log10((t + 1) / 25)))
def q_init(self, gamma, alpha, epsilon):
"""Initialize qlearning parameters"""
self.Q = defaultdict(float)
self.gamma = gamma
self.alpha = alpha
self.epsilon = epsilon
def state_init(self, high, low):
"""Initialize function approximation paramaters"""
self.num_buckets = (1, 1, 6, 3)
self.num_actions = self.action_space.n
self.state_bounds = list(zip(low, high))
self.state_bounds[1] = [-0.5, 0.5]
self.state_bounds[3] = [-math.radians(50), math.radians(50)]
def __init__(self, action_space,
high, low,
gamma=0.99, alpha=0.5, epsilon=0.5):
"""Initialize the agent"""
super(QlearningAgent, self).__init__(action_space)
self.q_init(gamma, alpha, epsilon)
self.state_init(high, low)
def act(self, state):
"""Take an action"""
if np.random.random()<self.epsilon:
i = np.random.randint(0, self.num_actions)
else:
s1 = self.state_to_bucket(state)
vals = [v for ((s, a), v) in self.Q.items() if s == s1]
if len(vals) <= 0:
i = np.random.randint(0, self.num_actions)
else:
i = np.argmax(vals)
return i
def learn(self, state, action, reward, next_state):
"""Learn with qlearning"""
s1, s2 = self.state_to_bucket(state), self.state_to_bucket(next_state)
self.Q[(s1, action)] += 0
for a in range(self.num_actions):
self.Q[(s2, a)] += 0
vals = [v for ((s, a), v) in self.Q.items() if s == s2]
max_q = max(vals)
td_target = reward + self.gamma * max_q
td_delta = td_target - self.Q[(s1, action)]
self.Q[(s1, action)] += self.alpha * td_delta
def state_to_bucket(self, state):
"""Function approximation"""
bucket_indice = []
for i in range(len(state)):
if state[i] <= self.state_bounds[i][0]:
bucket_index = 0
elif state[i] >= self.state_bounds[i][1]:
bucket_index = self.num_buckets[i] - 1
else:
# Mapping the state bounds to the bucket array
bound_width = self.state_bounds[i][1] - self.state_bounds[i][0]
offset = (self.num_buckets[i] - 1) * self.state_bounds[i][0] / bound_width
scaling = (self.num_buckets[i] - 1) / bound_width
bucket_index = int(round(scaling * state[i] - offset))
bucket_indice.append(bucket_index)
return tuple(bucket_indice)