-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathusing_jit.py
170 lines (141 loc) · 4.57 KB
/
using_jit.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
# coding:utf-8
from numba import jit, f8, i1
import numpy as np
import scipy.stats as st
@jit(f8[:](f8[:], f8[:]))
def Resampling(w_tilde, X10):
M = w_tilde.size
sum_of_w_tilde = np.zeros(M)
for i in range(M):
sum_of_w_tilde[i] = np.sum(w_tilde[:i + 1])
d = np.random.rand() / M # 0~1/Mの間の値をとる一様乱数
Arrow = np.arange(M) / M + d
X11 = np.array([])
for i in range(M):
if i == 0:
Resample = 0
for j in range(M):
if Arrow[j] < sum_of_w_tilde[i]:
Resample += 1
else:
break
else:
L0 = 0
for j in range(M):
if Arrow[j] <= sum_of_w_tilde[i - 1]:
L0 += 1
else:
break
L1 = 0
for j in range(M):
if Arrow[j] < sum_of_w_tilde[i]:
L1 += 1
else:
break
Resample = L1 - L0
if Resample == 0:
pass
else:
X11 = np.r_[X11, np.ones(Resample) * X10[i]]
if X11.size != M:
print(X10, X11)
raise ValueError('your X11\'s size is not valid!')
return X11
@jit(f8[:, :](f8[:], f8[:, :]))
def Resampling_2D(w_tilde, X10):
M = w_tilde.size
sum_of_w_tilde = np.zeros(M)
for i in range(M):
sum_of_w_tilde[i] = np.sum(w_tilde[:i + 1])
d = np.random.rand() / M # 0~1/Mの間の値をとる一様乱数
Arrow = np.arange(M) / M + d
X11 = np.array([])
for i in range(M):
if i == 0:
Resample = 0
for j in range(M):
if Arrow[j] < sum_of_w_tilde[i]:
Resample += 1
else:
break
else:
L0 = 0
for j in range(M):
if Arrow[j] <= sum_of_w_tilde[i - 1]:
L0 += 1
else:
break
L1 = 0
for j in range(M):
if Arrow[j] < sum_of_w_tilde[i]:
L1 += 1
else:
break
Resample = L1 - L0
if Resample == 0:
pass
else:
if X11.size == 0:
X11 = X10[:, i]
for k in range(Resample - 1):
X11 = np.c_[X11, X10[:, i]]
else:
for k in range(Resample):
X11 = np.c_[X11, X10[:, i]]
if X11.shape[0] != X10.shape[0]:
raise ValueError('your X11\'s size is not valid!')
return X11
@jit(f8(f8[:], f8[:]))
def cal_RMSE(xa, xb):
"""
xa, xbという2つのプロファイルの間のroot mean square errorを計算する。
xa, xbの各要素のどれか1つでもnanであれば、nanが返ってくる。
"""
if xa.size != xb.size:
raise Exception('error! xa\'s size and xb\'s are not different!')
if np.sum(xa) == 0 or np.sum(xb) == 0:
# xaとxbのどちらか一方が全部0であれば、その時刻ではrmseはnanにする。
return np.nan
else:
err = xa - xb
return np.sqrt(np.sum(err ** 2) / err.size)
@jit(f8[:](f8[:, :], f8[:, :]))
def cal_RMSE_2D(xa, xb):
"""
# xaが2次元(J, N)の時、j=1, ..., J毎にRMSEを計算して、長さJの配列を返す。
"""
if xa.shape != xb.shape:
raise Exception('error! xa\'s size and xb\'s are not different!')
J = xa.shape[0]
rmses = np.zeros(J)
for j in range(J):
rmses[j] = cal_RMSE(xa[j], xb[j])
return rmses
@jit(f8[:, :](f8[:], f8[:]))
def cal_covmat(err1, err2):
"""
誤差から誤差共分散行列を作成する。
err1.size = 40でテストした結果、
err1 * err2[:, None]よりもこの実装のほうが僅かに高速でした。
http://yukara-13.hatenablog.com/entry/2014/01/24/131640
http://lv4.hateblo.jp/entry/2014/07/23/132849
http://emoson.hateblo.jp/entry/2014/10/26/133736
"""
N = err1.size
Cov = np.zeros((N, N))
for i in range(N):
for j in range(N):
Cov[i, j] = err1[i] * err2[j]
return Cov
@jit(f8[:, :](f8[:, :], f8[:], f8[:], i1, i1))
def likelihood_(HXf, y, diagR, p, m):
"""
モデルグリッド毎に尤度を計算
(n(観測があるグリッドの個数), m(アンサンブル数))の形の配列を返す
HXf.shape == (p, m)
y.size == p
"""
lkh_mat = np.zeros((p, m))
for i in range(p):
lkh_mat[i] = st.norm.pdf(HXf[i] - y[i], scale=np.sqrt(diagR[i]))
return lkh_mat