質問編集履歴
2
読みやすくしました。
title
CHANGED
|
File without changes
|
body
CHANGED
|
@@ -45,7 +45,8 @@
|
|
|
45
45
|
renderer.draw_markers(
|
|
46
46
|
TypeError: must be real number, not str
|
|
47
47
|
|
|
48
|
-
----------------------------------------------------------------------------------------------------
|
|
48
|
+
----------------------------------------------------------------------------------------------------
|
|
49
|
+
```python
|
|
49
50
|
#!/usr/bin/env python3
|
|
50
51
|
#coding:utf-8
|
|
51
52
|
#
|
|
@@ -226,4 +227,5 @@
|
|
|
226
227
|
#
|
|
227
228
|
# Draw
|
|
228
229
|
#
|
|
229
|
-
plt.show()
|
|
230
|
+
plt.show()
|
|
231
|
+
```
|
1
report7.py追加しました。お願いします
title
CHANGED
|
File without changes
|
body
CHANGED
|
@@ -43,4 +43,187 @@
|
|
|
43
43
|
return draw(artist, renderer, *args, **kwargs)
|
|
44
44
|
File "/home/g2021048/.local/lib/python3.8/site-packages/matplotlib/collections.py", line 406, in draw
|
|
45
45
|
renderer.draw_markers(
|
|
46
|
-
TypeError: must be real number, not str
|
|
46
|
+
TypeError: must be real number, not str
|
|
47
|
+
|
|
48
|
+
------------------------------------------------------------------------------------------------------
|
|
49
|
+
#!/usr/bin/env python3
|
|
50
|
+
#coding:utf-8
|
|
51
|
+
#
|
|
52
|
+
# パターンが2次元連続変数で
|
|
53
|
+
# 多変量正規分布(Multivariate Normal Distribution) でモデル化した場合の
|
|
54
|
+
# 識別境界面
|
|
55
|
+
#
|
|
56
|
+
from mpl_toolkits.mplot3d import Axes3D
|
|
57
|
+
import numpy as np
|
|
58
|
+
import matplotlib.pyplot as plt
|
|
59
|
+
import matplotlib.tri as tri
|
|
60
|
+
import matplotlib.cm as cm
|
|
61
|
+
import functools
|
|
62
|
+
import scipy.stats as st
|
|
63
|
+
import scipy.special as sp
|
|
64
|
+
import math
|
|
65
|
+
import sys
|
|
66
|
+
from mpl_toolkits.axes_grid1 import make_axes_locatable
|
|
67
|
+
|
|
68
|
+
|
|
69
|
+
class Norm2D:
|
|
70
|
+
"""
|
|
71
|
+
2変量正規分布 (2 dimensional normal distribution)
|
|
72
|
+
"""
|
|
73
|
+
def __init__(self, X):
|
|
74
|
+
self.__X = np.copy(X)
|
|
75
|
+
self.__Mu = np.mean(self.__X, axis=0)
|
|
76
|
+
self.__Sigma = np.cov(self.__X, rowvar=False, bias=True)
|
|
77
|
+
self.__dim = len(self.__Mu)
|
|
78
|
+
self.__N = len(self.__X)
|
|
79
|
+
|
|
80
|
+
def get_N(self):
|
|
81
|
+
""" 学習データ数の取得 """
|
|
82
|
+
return self.__N
|
|
83
|
+
|
|
84
|
+
def get_X(self):
|
|
85
|
+
""" 学習データの取得 """
|
|
86
|
+
return self.__X
|
|
87
|
+
|
|
88
|
+
def get_param(self):
|
|
89
|
+
""" 正規分布パラメタの取得 """
|
|
90
|
+
return self.__Mu, self.__Sigma
|
|
91
|
+
|
|
92
|
+
def pdf(self, x):
|
|
93
|
+
""" 与えられた分布の(x)における確率密度値を求める"""
|
|
94
|
+
return st.multivariate_normal.pdf(x, mean=self.__Mu, cov=self.__Sigma)
|
|
95
|
+
|
|
96
|
+
def sampling(self, N):
|
|
97
|
+
""" 与えられた分布に従ってN点サンプリングする"""
|
|
98
|
+
return st.multivariate_normal.rvs(mean=self.__Mu, cov=self.__Sigma, size=N)
|
|
99
|
+
|
|
100
|
+
def this_likelihood(self, x):
|
|
101
|
+
""" パターンxが与えられた時の与えられた分布での尤度を求める """
|
|
102
|
+
L = 1.0
|
|
103
|
+
for n in range(len(x)):
|
|
104
|
+
L *= self.pdf(x[n])
|
|
105
|
+
return L
|
|
106
|
+
|
|
107
|
+
def this_log_likelihood(self, x):
|
|
108
|
+
""" パターンxが与えられた時の与えられた分布での対数尤度を求める """
|
|
109
|
+
logL = 0.0
|
|
110
|
+
for n in range(len(x)):
|
|
111
|
+
logL += log(self.pdf(x[n]))
|
|
112
|
+
return logL
|
|
113
|
+
|
|
114
|
+
|
|
115
|
+
# 例題2
|
|
116
|
+
# x(ω1) = {(0, 4√2), (4, 4√2), (2, 4√2-1), (2, 4√2+1)}
|
|
117
|
+
# x(ω2) = {(0, 0), (4, 0), (2, -1), (2, 1), (2(1-√3), 0), (2(1+√3), 0), (2, -√3), (2, √3)}
|
|
118
|
+
"""
|
|
119
|
+
x1 = np.array([[0.0, 4.0*np.sqrt(2.0)],
|
|
120
|
+
[4.0, 4.0*np.sqrt(2.0)],
|
|
121
|
+
[2.0, 4.0*np.sqrt(2.0)-1.0],
|
|
122
|
+
[2.0, 4.0*np.sqrt(2.0)+1.0]])
|
|
123
|
+
x2 = np.array([[0.0, 0.0],
|
|
124
|
+
[4.0, 0.0],
|
|
125
|
+
[2.0,-1.0],
|
|
126
|
+
[2.0, 1.0],
|
|
127
|
+
[2.0*(1.0-np.sqrt(3.0)), 0],
|
|
128
|
+
[2.0*(1.0+np.sqrt(3.0)), 0],
|
|
129
|
+
[2.0,-np.sqrt(3.0)],
|
|
130
|
+
[2.0, np.sqrt(3.0)]])
|
|
131
|
+
# Set the drow range
|
|
132
|
+
x = y = np.arange(-2, 18, 0.1)
|
|
133
|
+
"""
|
|
134
|
+
|
|
135
|
+
#
|
|
136
|
+
# 例題3
|
|
137
|
+
# x(ω1) = {(-2, 0), (0, 1), (0, -1), (2, 0)}
|
|
138
|
+
# x(ω2) = {(3, 2), (4, -1), (6, 1), (7, -2)}
|
|
139
|
+
x1 = np.array([[-2.0, 0.0],
|
|
140
|
+
[0.0, 1.0],
|
|
141
|
+
[0.0,-1.0],
|
|
142
|
+
[2.0, 0.0]])
|
|
143
|
+
x2 = np.array([[3.0, 2.0],
|
|
144
|
+
[4.0,-1.0],
|
|
145
|
+
[6.0, 1.0],
|
|
146
|
+
[7.0,-2.0]])
|
|
147
|
+
# Set the drow range
|
|
148
|
+
x = y = np.arange(-5, 10, 0.1)
|
|
149
|
+
|
|
150
|
+
#
|
|
151
|
+
# Create Distributions
|
|
152
|
+
#
|
|
153
|
+
dist1 = Norm2D(x1)
|
|
154
|
+
dist2 = Norm2D(x2)
|
|
155
|
+
|
|
156
|
+
#
|
|
157
|
+
# Show Mu and Sigma
|
|
158
|
+
#
|
|
159
|
+
Mu1, Sigma1 = dist1.get_param()
|
|
160
|
+
print('Mu1=', Mu1);
|
|
161
|
+
print('Sigma1=', Sigma1)
|
|
162
|
+
|
|
163
|
+
Mu2, Sigma2 = dist2.get_param()
|
|
164
|
+
print('Mu2=', Mu2)
|
|
165
|
+
print('Sigma2=', Sigma2)
|
|
166
|
+
|
|
167
|
+
|
|
168
|
+
#
|
|
169
|
+
# Create the drawing mesh
|
|
170
|
+
#
|
|
171
|
+
X, Y = np.meshgrid(x, y)
|
|
172
|
+
pos = np.dstack((X,Y))
|
|
173
|
+
|
|
174
|
+
#
|
|
175
|
+
# Calc. pdf
|
|
176
|
+
#
|
|
177
|
+
Z1 = dist1.pdf(pos)
|
|
178
|
+
Z2 = dist2.pdf(pos)
|
|
179
|
+
|
|
180
|
+
Z = np.log(np.fmax(Z1, Z2)) # Logarithmic expression
|
|
181
|
+
Zdiff = Z1 - Z2
|
|
182
|
+
|
|
183
|
+
maxZ = np.max(Z)
|
|
184
|
+
minZ = np.min(Z)
|
|
185
|
+
#print('max Z=', maxZ)
|
|
186
|
+
#print('min Z=', minZ)
|
|
187
|
+
|
|
188
|
+
minZ = -20 # Clip the lowest Z value
|
|
189
|
+
|
|
190
|
+
|
|
191
|
+
#
|
|
192
|
+
# Init. a graph
|
|
193
|
+
#
|
|
194
|
+
fig = plt.figure(figsize=(7,5))
|
|
195
|
+
|
|
196
|
+
ax = fig.add_subplot(111, aspect='equal')
|
|
197
|
+
ax.grid()
|
|
198
|
+
plt.xlabel('x')
|
|
199
|
+
plt.ylabel('y')
|
|
200
|
+
|
|
201
|
+
|
|
202
|
+
#
|
|
203
|
+
# draw the contour
|
|
204
|
+
#
|
|
205
|
+
levels = np.linspace(minZ, maxZ, 50)
|
|
206
|
+
cs = ax.contourf(X, Y, Z, levels=levels, cmap=cm.inferno, extend='both')
|
|
207
|
+
|
|
208
|
+
#
|
|
209
|
+
# draw the boundary
|
|
210
|
+
#
|
|
211
|
+
ax.contour(X, Y, Zdiff, levels=[-1.0e-300, 1.0e-300], colors='r', linestyles='-')
|
|
212
|
+
|
|
213
|
+
#
|
|
214
|
+
# data
|
|
215
|
+
#
|
|
216
|
+
ax.scatter(dist1.get_X()[:,0], dist1.get_X()[:,1], s=40, c='red', marker='o', alpha=0.5, linewidths='2')
|
|
217
|
+
ax.scatter(dist2.get_X()[:,0], dist2.get_X()[:,1], s=40, c='blue', marker='o', alpha=0.5, linewidths='2')
|
|
218
|
+
|
|
219
|
+
#
|
|
220
|
+
# draw a colorbar
|
|
221
|
+
#
|
|
222
|
+
divider = make_axes_locatable(ax) # create the same size divider as the contour graph
|
|
223
|
+
cax = divider.append_axes('right', size='5%', pad=0.1) # append 5% space to the right-end
|
|
224
|
+
fig.colorbar(cs, cax=cax, format='%.1f') # show a colorbar
|
|
225
|
+
|
|
226
|
+
#
|
|
227
|
+
# Draw
|
|
228
|
+
#
|
|
229
|
+
plt.show()
|