-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy path3_3.py
141 lines (112 loc) · 3.64 KB
/
3_3.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
import numpy as np
import matplotlib.pyplot as plt
from GridWorld_env import *
# from nSARSA import *
# from Q_Learning import *
# from n_step_bt import *
# # from On_MC import *
from Policy_Iteration_with_MC import *
from Policy_Iteration import *
import seaborn as sns
SEED = 184
ba_s=[None for _ in range(3)]
br_s=[0 for _ in range(3)]
def mean_without_outliers(data, k=1.5):
"""
Calculate the mean of the data while ignoring outliers using Tukey's fences.
Parameters:
- data (numpy.ndarray or list): Input data.
- k (float): Tukey's fences constant. Typically set to 1.5.
Returns:
- float: Mean without outliers.
"""
q1 = np.percentile(data, 25)
q3 = np.percentile(data, 75)
iqr = q3 - q1
lower_bound = q1 - k * iqr
upper_bound = q3 + k * iqr
filtered_data = [x for x in data if lower_bound <= x <= upper_bound]
return np.mean(filtered_data)
def get_optimal_policy(q_table):
new_q_table = np.zeros((q_table.shape[0], q_table.shape[1]))
for i in range(len(q_table)):
index = np.argmax(q_table[i])
new_q_table[i][index] = 1
return new_q_table
def draw_curves(values):
cmap = plt.get_cmap('tab10')
random_colors = [cmap(i) for i in np.linspace(0, 1, 10)]
plt.figure()
flag =True
for _ in values:
vals1=_[0]
label1=_[1]
mean_value = np.mean(vals1, axis=0)
std_value = np.std(vals1, axis=0)
upper_bound = mean_value + 1.96 * std_value / np.sqrt(len(vals1[0]))
lower_bound = mean_value - 1.96 * std_value / np.sqrt(len(vals1[0]))
plt.plot(mean_value, label=label1, color=random_colors[values.index(_)])
plt.fill_between(
np.arange(len(mean_value)),
lower_bound,
upper_bound,
alpha=0.5,
label=f"Confidence Interval for {label1}",
color=random_colors[values.index(_)]
)
flag=False
plt.legend()
plt.show()
rewards=[]
gwe = GridWorldEnv(size=6)
np.random.seed(SEED)
random.seed(SEED)
start=0.1
list1=[[[],'number of Evaluation_Improvment before finding optimal policy PI'],[[],'number of Evaluation_Improvment before finding optimal policy PI&MC-method1']]
for i in range(10):
print(f"Agent{i+1}")
gwe.reset()
agent = Policy_Iteration(gwe, 0.9, 0.1)
rs, q_table,count = agent.fit_episode_generating()
list1[0][0].append(rs)
# list1[0][1]=np.mean(list1[0][0])
if mean_without_outliers(rs)>br_s[0]:
br_s[0]=mean_without_outliers(rs)
ba_s[0]=agent
for i in range(10):
print(f"Agent{i+1}")
gwe.reset()
agent = Policy_IterationWMC(gwe, 0.9, 0.1)
rs, q_table,count = agent.fit()
list1[1][0].append(rs)
# list1[1][1]=np.mean(list1[1][0])
if mean_without_outliers(rs)>br_s[1]:
br_s[1]=mean_without_outliers(rs)
ba_s[1]=agent
for i in range(10):
print(f"Agent{i+1}")
gwe.reset()
agent = Policy_IterationWMC(gwe, 0.9, 0.1)
rs, q_table,count = agent.fit_2()
list1[2][0].append(rs)
# list1[2][1]=np.mean(list1[2][0])
if mean_without_outliers(rs)>br_s[2]:
br_s[2]=mean_without_outliers(rs)
ba_s[2]=agent
draw_curves(list1)
for z in range(3):
reward=[]
ep_len=[]
for _ in range(10):
gwe.reset()
gwe.set_render()
ep=ba_s[z].generate_episode()
ep_len=len(ep)
rew=0
for x in ep:
s,a,r=x
rew+=r
reward.append(rew)
print('i assumed that entering fianl state is equal to getting 1000 reward')
print(f'mean reward for agent {z}={np.mean(reward)}')
print(f'mean episode length for agent {z}={np.mean(ep_len)}')