foscat 3.0.9__py3-none-any.whl → 3.6.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- foscat/CNN.py +151 -0
- foscat/CircSpline.py +102 -34
- foscat/FoCUS.py +2363 -1052
- foscat/GCNN.py +239 -0
- foscat/Softmax.py +29 -20
- foscat/Spline1D.py +86 -36
- foscat/Synthesis.py +335 -262
- foscat/alm.py +690 -0
- foscat/alm_tools.py +11 -0
- foscat/backend.py +933 -588
- foscat/backend_tens.py +63 -0
- foscat/loss_backend_tens.py +48 -38
- foscat/loss_backend_torch.py +35 -41
- foscat/scat.py +1639 -1015
- foscat/scat1D.py +1256 -774
- foscat/scat2D.py +9 -7
- foscat/scat_cov.py +3067 -1541
- foscat/scat_cov1D.py +11 -1467
- foscat/scat_cov2D.py +9 -7
- foscat/scat_cov_map.py +77 -51
- foscat/scat_cov_map2D.py +79 -49
- foscat-3.6.0.dist-info/LICENCE +13 -0
- foscat-3.6.0.dist-info/METADATA +184 -0
- foscat-3.6.0.dist-info/RECORD +27 -0
- {foscat-3.0.9.dist-info → foscat-3.6.0.dist-info}/WHEEL +1 -1
- foscat/GetGPUinfo.py +0 -36
- foscat-3.0.9.dist-info/METADATA +0 -23
- foscat-3.0.9.dist-info/RECORD +0 -22
- {foscat-3.0.9.dist-info → foscat-3.6.0.dist-info}/top_level.txt +0 -0
foscat/scat.py
CHANGED
|
@@ -1,412 +1,602 @@
|
|
|
1
|
-
import foscat.FoCUS as FOC
|
|
2
|
-
import numpy as np
|
|
3
|
-
import tensorflow as tf
|
|
4
1
|
import pickle
|
|
5
|
-
import
|
|
2
|
+
import sys
|
|
3
|
+
|
|
6
4
|
import healpy as hp
|
|
7
|
-
|
|
5
|
+
import numpy as np
|
|
6
|
+
|
|
7
|
+
import foscat.backend as bk
|
|
8
|
+
import foscat.FoCUS as FOC
|
|
9
|
+
|
|
10
|
+
# Vérifier si TensorFlow est importé et défini
|
|
11
|
+
tf_defined = "tensorflow" in sys.modules
|
|
12
|
+
|
|
13
|
+
if tf_defined:
|
|
14
|
+
import tensorflow as tf
|
|
15
|
+
|
|
16
|
+
tf_function = (
|
|
17
|
+
tf.function
|
|
18
|
+
) # Facultatif : si vous voulez utiliser TensorFlow dans ce script
|
|
19
|
+
else:
|
|
20
|
+
|
|
21
|
+
def tf_function(func):
|
|
22
|
+
return func
|
|
23
|
+
|
|
24
|
+
|
|
8
25
|
def read(filename):
|
|
9
|
-
thescat=scat(1,1,1,1,1,[0],[0])
|
|
26
|
+
thescat = scat(1, 1, 1, 1, 1, [0], [0])
|
|
10
27
|
return thescat.read(filename)
|
|
11
|
-
|
|
28
|
+
|
|
29
|
+
|
|
12
30
|
class scat:
|
|
13
|
-
def __init__(self,p00,s0,s1,s2,s2l,j1,j2,cross=False,backend=None):
|
|
14
|
-
self.bk_type=
|
|
15
|
-
self.P00=p00
|
|
16
|
-
self.S0=s0
|
|
17
|
-
self.S1=s1
|
|
18
|
-
self.S2=s2
|
|
19
|
-
self.S2L=s2l
|
|
20
|
-
self.j1=j1
|
|
21
|
-
self.j2=j2
|
|
22
|
-
self.cross=cross
|
|
23
|
-
self.backend=backend
|
|
24
|
-
|
|
25
|
-
def set_bk_type(self,bk_type):
|
|
26
|
-
self.bk_type=bk_type
|
|
27
|
-
|
|
31
|
+
def __init__(self, p00, s0, s1, s2, s2l, j1, j2, cross=False, backend=None):
|
|
32
|
+
self.bk_type = "SCAT"
|
|
33
|
+
self.P00 = p00
|
|
34
|
+
self.S0 = s0
|
|
35
|
+
self.S1 = s1
|
|
36
|
+
self.S2 = s2
|
|
37
|
+
self.S2L = s2l
|
|
38
|
+
self.j1 = j1
|
|
39
|
+
self.j2 = j2
|
|
40
|
+
self.cross = cross
|
|
41
|
+
self.backend = backend
|
|
42
|
+
|
|
43
|
+
def set_bk_type(self, bk_type):
|
|
44
|
+
self.bk_type = bk_type
|
|
45
|
+
|
|
28
46
|
def get_j_idx(self):
|
|
29
|
-
return self.j1,self.j2
|
|
30
|
-
|
|
47
|
+
return self.j1, self.j2
|
|
48
|
+
|
|
31
49
|
def get_S0(self):
|
|
32
|
-
return
|
|
50
|
+
return self.S0
|
|
33
51
|
|
|
34
52
|
def get_S1(self):
|
|
35
|
-
return
|
|
36
|
-
|
|
53
|
+
return self.S1
|
|
54
|
+
|
|
37
55
|
def get_S2(self):
|
|
38
|
-
return
|
|
56
|
+
return self.S2
|
|
39
57
|
|
|
40
58
|
def get_S2L(self):
|
|
41
|
-
return
|
|
59
|
+
return self.S2L
|
|
42
60
|
|
|
43
61
|
def get_P00(self):
|
|
44
|
-
return
|
|
62
|
+
return self.P00
|
|
45
63
|
|
|
46
64
|
def reset_P00(self):
|
|
47
|
-
self.P00=0*self.P00
|
|
65
|
+
self.P00 = 0 * self.P00
|
|
48
66
|
|
|
49
67
|
def constant(self):
|
|
50
|
-
return scat(
|
|
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
|
-
def
|
|
78
|
-
|
|
79
|
-
return x
|
|
80
|
-
|
|
81
|
-
|
|
82
|
-
|
|
83
|
-
|
|
84
|
-
|
|
85
|
-
|
|
86
|
-
|
|
87
|
-
|
|
88
|
-
|
|
89
|
-
|
|
90
|
-
|
|
91
|
-
|
|
92
|
-
|
|
93
|
-
|
|
94
|
-
|
|
95
|
-
return
|
|
96
|
-
|
|
97
|
-
|
|
98
|
-
|
|
99
|
-
|
|
100
|
-
|
|
101
|
-
|
|
102
|
-
|
|
103
|
-
self.backend.
|
|
104
|
-
|
|
105
|
-
|
|
106
|
-
|
|
107
|
-
|
|
108
|
-
|
|
109
|
-
|
|
68
|
+
return scat(
|
|
69
|
+
self.backend.constant(self.P00),
|
|
70
|
+
self.backend.constant(self.S0),
|
|
71
|
+
self.backend.constant(self.S1),
|
|
72
|
+
self.backend.constant(self.S2),
|
|
73
|
+
self.backend.constant(self.S2L),
|
|
74
|
+
self.j1,
|
|
75
|
+
self.j2,
|
|
76
|
+
backend=self.backend,
|
|
77
|
+
)
|
|
78
|
+
|
|
79
|
+
def domult(self, x, y):
|
|
80
|
+
try:
|
|
81
|
+
return x * y
|
|
82
|
+
except:
|
|
83
|
+
if x.dtype == y.dtype:
|
|
84
|
+
return x * y
|
|
85
|
+
if self.backend.bk_is_complex(x):
|
|
86
|
+
|
|
87
|
+
return self.backend.bk_complex(
|
|
88
|
+
self.backend.bk_real(x) * y, self.backend.bk_imag(x) * y
|
|
89
|
+
)
|
|
90
|
+
else:
|
|
91
|
+
return self.backend.bk_complex(
|
|
92
|
+
self.backend.bk_real(y) * x, self.backend.bk_imag(y) * x
|
|
93
|
+
)
|
|
94
|
+
|
|
95
|
+
def dodiv(self, x, y):
|
|
96
|
+
try:
|
|
97
|
+
return x / y
|
|
98
|
+
except:
|
|
99
|
+
if x.dtype == y.dtype:
|
|
100
|
+
return x / y
|
|
101
|
+
if self.backend.bk_is_complex(x):
|
|
102
|
+
|
|
103
|
+
return self.backend.bk_complex(
|
|
104
|
+
self.backend.bk_real(x) / y, self.backend.bk_imag(x) / y
|
|
105
|
+
)
|
|
106
|
+
else:
|
|
107
|
+
return self.backend.bk_complex(
|
|
108
|
+
x / self.backend.bk_real(y), x / self.backend.bk_imag(y)
|
|
109
|
+
)
|
|
110
|
+
|
|
111
|
+
def domin(self, x, y):
|
|
112
|
+
try:
|
|
113
|
+
return x - y
|
|
114
|
+
except:
|
|
115
|
+
if x.dtype == y.dtype:
|
|
116
|
+
return x - y
|
|
117
|
+
|
|
118
|
+
if self.backend.bk_is_complex(x):
|
|
119
|
+
|
|
120
|
+
return self.backend.bk_complex(
|
|
121
|
+
self.backend.bk_real(x) - y, self.backend.bk_imag(x) - y
|
|
122
|
+
)
|
|
123
|
+
else:
|
|
124
|
+
return self.backend.bk_complex(
|
|
125
|
+
x - self.backend.bk_real(y), x - self.backend.bk_imag(y)
|
|
126
|
+
)
|
|
127
|
+
|
|
128
|
+
def doadd(self, x, y):
|
|
129
|
+
try:
|
|
130
|
+
return x + y
|
|
131
|
+
except:
|
|
132
|
+
if x.dtype == y.dtype:
|
|
133
|
+
return x + y
|
|
134
|
+
if self.backend.bk_is_complex(x):
|
|
135
|
+
|
|
136
|
+
return self.backend.bk_complex(
|
|
137
|
+
self.backend.bk_real(x) + y, self.backend.bk_imag(x) + y
|
|
138
|
+
)
|
|
139
|
+
else:
|
|
140
|
+
return self.backend.bk_complex(
|
|
141
|
+
x + self.backend.bk_real(y), x + self.backend.bk_imag(y)
|
|
142
|
+
)
|
|
143
|
+
|
|
144
|
+
def __add__(self, other):
|
|
145
|
+
assert (
|
|
146
|
+
isinstance(other, float)
|
|
147
|
+
or isinstance(other, np.float32)
|
|
148
|
+
or isinstance(other, int)
|
|
149
|
+
or isinstance(other, bool)
|
|
150
|
+
or isinstance(other, scat)
|
|
151
|
+
)
|
|
152
|
+
|
|
110
153
|
if isinstance(other, scat):
|
|
111
|
-
return scat(
|
|
112
|
-
|
|
113
|
-
|
|
114
|
-
|
|
115
|
-
|
|
116
|
-
|
|
154
|
+
return scat(
|
|
155
|
+
self.doadd(self.P00, other.P00),
|
|
156
|
+
self.doadd(self.S0, other.S0),
|
|
157
|
+
self.doadd(self.S1, other.S1),
|
|
158
|
+
self.doadd(self.S2, other.S2),
|
|
159
|
+
self.doadd(self.S2L, other.S2L),
|
|
160
|
+
self.j1,
|
|
161
|
+
self.j2,
|
|
162
|
+
backend=self.backend,
|
|
163
|
+
)
|
|
117
164
|
else:
|
|
118
|
-
return scat(
|
|
119
|
-
|
|
120
|
-
|
|
121
|
-
|
|
122
|
-
|
|
123
|
-
|
|
124
|
-
|
|
125
|
-
|
|
165
|
+
return scat(
|
|
166
|
+
(self.P00 + other),
|
|
167
|
+
(self.S0 + other),
|
|
168
|
+
(self.S1 + other),
|
|
169
|
+
(self.S2 + other),
|
|
170
|
+
(self.S2L + other),
|
|
171
|
+
self.j1,
|
|
172
|
+
self.j2,
|
|
173
|
+
backend=self.backend,
|
|
174
|
+
)
|
|
175
|
+
|
|
176
|
+
def toreal(self, value):
|
|
126
177
|
if value is None:
|
|
127
178
|
return None
|
|
128
|
-
|
|
179
|
+
|
|
129
180
|
return self.backend.bk_real(value)
|
|
130
181
|
|
|
131
|
-
def addcomplex(self,value,amp):
|
|
182
|
+
def addcomplex(self, value, amp):
|
|
132
183
|
if value is None:
|
|
133
184
|
return None
|
|
134
|
-
|
|
135
|
-
return self.backend.bk_complex(value,amp*value)
|
|
136
|
-
|
|
137
|
-
def add_complex(self,amp):
|
|
138
|
-
return scat(
|
|
139
|
-
|
|
140
|
-
|
|
141
|
-
|
|
142
|
-
|
|
143
|
-
|
|
144
|
-
|
|
185
|
+
|
|
186
|
+
return self.backend.bk_complex(value, amp * value)
|
|
187
|
+
|
|
188
|
+
def add_complex(self, amp):
|
|
189
|
+
return scat(
|
|
190
|
+
self.addcomplex(self.P00, amp),
|
|
191
|
+
self.addcomplex(self.S0, amp),
|
|
192
|
+
self.addcomplex(self.S1, amp),
|
|
193
|
+
self.addcomplex(self.S2, amp),
|
|
194
|
+
self.addcomplex(self.S2L, amp),
|
|
195
|
+
self.j1,
|
|
196
|
+
self.j2,
|
|
197
|
+
backend=self.backend,
|
|
198
|
+
)
|
|
199
|
+
|
|
145
200
|
def real(self):
|
|
146
|
-
return scat(
|
|
147
|
-
|
|
148
|
-
|
|
149
|
-
|
|
150
|
-
|
|
151
|
-
|
|
152
|
-
|
|
153
|
-
|
|
201
|
+
return scat(
|
|
202
|
+
self.toreal(self.P00),
|
|
203
|
+
self.toreal(self.S0),
|
|
204
|
+
self.toreal(self.S1),
|
|
205
|
+
self.toreal(self.S2),
|
|
206
|
+
self.toreal(self.S2L),
|
|
207
|
+
self.j1,
|
|
208
|
+
self.j2,
|
|
209
|
+
backend=self.backend,
|
|
210
|
+
)
|
|
211
|
+
|
|
212
|
+
def __radd__(self, other):
|
|
154
213
|
return self.__add__(other)
|
|
155
214
|
|
|
156
|
-
def __truediv__(self,other):
|
|
157
|
-
assert
|
|
158
|
-
isinstance(other,
|
|
159
|
-
|
|
215
|
+
def __truediv__(self, other):
|
|
216
|
+
assert (
|
|
217
|
+
isinstance(other, float)
|
|
218
|
+
or isinstance(other, np.float32)
|
|
219
|
+
or isinstance(other, int)
|
|
220
|
+
or isinstance(other, bool)
|
|
221
|
+
or isinstance(other, scat)
|
|
222
|
+
)
|
|
223
|
+
|
|
160
224
|
if isinstance(other, scat):
|
|
161
|
-
return scat(
|
|
162
|
-
|
|
163
|
-
|
|
164
|
-
|
|
165
|
-
|
|
166
|
-
|
|
225
|
+
return scat(
|
|
226
|
+
self.dodiv(self.P00, other.P00),
|
|
227
|
+
self.dodiv(self.S0, other.S0),
|
|
228
|
+
self.dodiv(self.S1, other.S1),
|
|
229
|
+
self.dodiv(self.S2, other.S2),
|
|
230
|
+
self.dodiv(self.S2L, other.S2L),
|
|
231
|
+
self.j1,
|
|
232
|
+
self.j2,
|
|
233
|
+
backend=self.backend,
|
|
234
|
+
)
|
|
167
235
|
else:
|
|
168
|
-
return scat(
|
|
169
|
-
|
|
170
|
-
|
|
171
|
-
|
|
172
|
-
|
|
173
|
-
|
|
174
|
-
|
|
175
|
-
|
|
176
|
-
|
|
177
|
-
|
|
178
|
-
|
|
179
|
-
|
|
236
|
+
return scat(
|
|
237
|
+
(self.P00 / other),
|
|
238
|
+
(self.S0 / other),
|
|
239
|
+
(self.S1 / other),
|
|
240
|
+
(self.S2 / other),
|
|
241
|
+
(self.S2L / other),
|
|
242
|
+
self.j1,
|
|
243
|
+
self.j2,
|
|
244
|
+
backend=self.backend,
|
|
245
|
+
)
|
|
246
|
+
|
|
247
|
+
def __rtruediv__(self, other):
|
|
248
|
+
assert (
|
|
249
|
+
isinstance(other, float)
|
|
250
|
+
or isinstance(other, np.float32)
|
|
251
|
+
or isinstance(other, int)
|
|
252
|
+
or isinstance(other, bool)
|
|
253
|
+
or isinstance(other, scat)
|
|
254
|
+
)
|
|
255
|
+
|
|
180
256
|
if isinstance(other, scat):
|
|
181
|
-
return scat(
|
|
182
|
-
|
|
183
|
-
|
|
184
|
-
|
|
185
|
-
|
|
186
|
-
|
|
257
|
+
return scat(
|
|
258
|
+
self.dodiv(other.P00, self.P00),
|
|
259
|
+
self.dodiv(other.S0, self.S0),
|
|
260
|
+
self.dodiv(other.S1, self.S1),
|
|
261
|
+
self.dodiv(other.S2, self.S2),
|
|
262
|
+
self.dodiv(other.S2L, self.S2L),
|
|
263
|
+
self.j1,
|
|
264
|
+
self.j2,
|
|
265
|
+
backend=self.backend,
|
|
266
|
+
)
|
|
187
267
|
else:
|
|
188
|
-
return scat(
|
|
189
|
-
|
|
190
|
-
|
|
191
|
-
|
|
192
|
-
|
|
193
|
-
|
|
194
|
-
|
|
195
|
-
|
|
196
|
-
|
|
197
|
-
|
|
198
|
-
|
|
268
|
+
return scat(
|
|
269
|
+
(other / self.P00),
|
|
270
|
+
(other / self.S0),
|
|
271
|
+
(other / self.S1),
|
|
272
|
+
(other / self.S2),
|
|
273
|
+
(other / self.S2L),
|
|
274
|
+
self.j1,
|
|
275
|
+
self.j2,
|
|
276
|
+
backend=self.backend,
|
|
277
|
+
)
|
|
278
|
+
|
|
279
|
+
def __sub__(self, other):
|
|
280
|
+
assert (
|
|
281
|
+
isinstance(other, float)
|
|
282
|
+
or isinstance(other, np.float32)
|
|
283
|
+
or isinstance(other, int)
|
|
284
|
+
or isinstance(other, bool)
|
|
285
|
+
or isinstance(other, scat)
|
|
286
|
+
)
|
|
287
|
+
|
|
199
288
|
if isinstance(other, scat):
|
|
200
|
-
return scat(
|
|
201
|
-
|
|
202
|
-
|
|
203
|
-
|
|
204
|
-
|
|
205
|
-
|
|
289
|
+
return scat(
|
|
290
|
+
self.domin(self.P00, other.P00),
|
|
291
|
+
self.domin(self.S0, other.S0),
|
|
292
|
+
self.domin(self.S1, other.S1),
|
|
293
|
+
self.domin(self.S2, other.S2),
|
|
294
|
+
self.domin(self.S2L, other.S2L),
|
|
295
|
+
self.j1,
|
|
296
|
+
self.j2,
|
|
297
|
+
backend=self.backend,
|
|
298
|
+
)
|
|
206
299
|
else:
|
|
207
|
-
return scat(
|
|
208
|
-
|
|
209
|
-
|
|
210
|
-
|
|
211
|
-
|
|
212
|
-
|
|
213
|
-
|
|
214
|
-
|
|
215
|
-
|
|
216
|
-
|
|
217
|
-
|
|
300
|
+
return scat(
|
|
301
|
+
(self.P00 - other),
|
|
302
|
+
(self.S0 - other),
|
|
303
|
+
(self.S1 - other),
|
|
304
|
+
(self.S2 - other),
|
|
305
|
+
(self.S2L - other),
|
|
306
|
+
self.j1,
|
|
307
|
+
self.j2,
|
|
308
|
+
backend=self.backend,
|
|
309
|
+
)
|
|
310
|
+
|
|
311
|
+
def __rsub__(self, other):
|
|
312
|
+
assert (
|
|
313
|
+
isinstance(other, float)
|
|
314
|
+
or isinstance(other, np.float32)
|
|
315
|
+
or isinstance(other, int)
|
|
316
|
+
or isinstance(other, bool)
|
|
317
|
+
or isinstance(other, scat)
|
|
318
|
+
)
|
|
319
|
+
|
|
218
320
|
if isinstance(other, scat):
|
|
219
|
-
return scat(
|
|
220
|
-
|
|
221
|
-
|
|
222
|
-
|
|
223
|
-
|
|
224
|
-
|
|
321
|
+
return scat(
|
|
322
|
+
self.domin(other.P00, self.P00),
|
|
323
|
+
self.domin(other.S0, self.S0),
|
|
324
|
+
self.domin(other.S1, self.S1),
|
|
325
|
+
self.domin(other.S2, self.S2),
|
|
326
|
+
self.domin(other.S2L, self.S2L),
|
|
327
|
+
self.j1,
|
|
328
|
+
self.j2,
|
|
329
|
+
backend=self.backend,
|
|
330
|
+
)
|
|
225
331
|
else:
|
|
226
|
-
return scat(
|
|
227
|
-
|
|
228
|
-
|
|
229
|
-
|
|
230
|
-
|
|
231
|
-
|
|
232
|
-
|
|
233
|
-
|
|
234
|
-
|
|
235
|
-
|
|
236
|
-
|
|
332
|
+
return scat(
|
|
333
|
+
(other - self.P00),
|
|
334
|
+
(other - self.S0),
|
|
335
|
+
(other - self.S1),
|
|
336
|
+
(other - self.S2),
|
|
337
|
+
(other - self.S2L),
|
|
338
|
+
self.j1,
|
|
339
|
+
self.j2,
|
|
340
|
+
backend=self.backend,
|
|
341
|
+
)
|
|
342
|
+
|
|
343
|
+
def __mul__(self, other):
|
|
344
|
+
assert (
|
|
345
|
+
isinstance(other, float)
|
|
346
|
+
or isinstance(other, np.float32)
|
|
347
|
+
or isinstance(other, int)
|
|
348
|
+
or isinstance(other, bool)
|
|
349
|
+
or isinstance(other, scat)
|
|
350
|
+
)
|
|
351
|
+
|
|
237
352
|
if isinstance(other, scat):
|
|
238
|
-
return scat(
|
|
239
|
-
|
|
240
|
-
|
|
241
|
-
|
|
242
|
-
|
|
243
|
-
|
|
353
|
+
return scat(
|
|
354
|
+
self.domult(self.P00, other.P00),
|
|
355
|
+
self.domult(self.S0, other.S0),
|
|
356
|
+
self.domult(self.S1, other.S1),
|
|
357
|
+
self.domult(self.S2, other.S2),
|
|
358
|
+
self.domult(self.S2L, other.S2L),
|
|
359
|
+
self.j1,
|
|
360
|
+
self.j2,
|
|
361
|
+
backend=self.backend,
|
|
362
|
+
)
|
|
244
363
|
else:
|
|
245
|
-
return scat(
|
|
246
|
-
|
|
247
|
-
|
|
248
|
-
|
|
249
|
-
|
|
250
|
-
|
|
364
|
+
return scat(
|
|
365
|
+
(self.P00 * other),
|
|
366
|
+
(self.S0 * other),
|
|
367
|
+
(self.S1 * other),
|
|
368
|
+
(self.S2 * other),
|
|
369
|
+
(self.S2L * other),
|
|
370
|
+
self.j1,
|
|
371
|
+
self.j2,
|
|
372
|
+
backend=self.backend,
|
|
373
|
+
)
|
|
374
|
+
|
|
251
375
|
def relu(self):
|
|
252
|
-
return scat(
|
|
253
|
-
|
|
254
|
-
|
|
255
|
-
|
|
256
|
-
|
|
257
|
-
|
|
258
|
-
|
|
259
|
-
|
|
260
|
-
|
|
261
|
-
|
|
262
|
-
|
|
263
|
-
|
|
376
|
+
return scat(
|
|
377
|
+
self.backend.bk_relu(self.P00),
|
|
378
|
+
self.backend.bk_relu(self.S0),
|
|
379
|
+
self.backend.bk_relu(self.S1),
|
|
380
|
+
self.backend.bk_relu(self.S2),
|
|
381
|
+
self.backend.bk_relu(self.S2L),
|
|
382
|
+
self.j1,
|
|
383
|
+
self.j2,
|
|
384
|
+
backend=self.backend,
|
|
385
|
+
)
|
|
386
|
+
|
|
387
|
+
def __rmul__(self, other):
|
|
388
|
+
assert (
|
|
389
|
+
isinstance(other, float)
|
|
390
|
+
or isinstance(other, np.float32)
|
|
391
|
+
or isinstance(other, int)
|
|
392
|
+
or isinstance(other, bool)
|
|
393
|
+
or isinstance(other, scat)
|
|
394
|
+
)
|
|
395
|
+
|
|
264
396
|
if isinstance(other, scat):
|
|
265
|
-
return scat(
|
|
266
|
-
|
|
267
|
-
|
|
268
|
-
|
|
269
|
-
|
|
270
|
-
|
|
397
|
+
return scat(
|
|
398
|
+
self.domult(self.P00, other.P00),
|
|
399
|
+
self.domult(self.S0, other.S0),
|
|
400
|
+
self.domult(self.S1, other.S1),
|
|
401
|
+
self.domult(self.S2, other.S2),
|
|
402
|
+
self.domult(self.S2L, other.S2L),
|
|
403
|
+
self.j1,
|
|
404
|
+
self.j2,
|
|
405
|
+
backend=self.backend,
|
|
406
|
+
)
|
|
271
407
|
else:
|
|
272
|
-
return scat(
|
|
273
|
-
|
|
274
|
-
|
|
275
|
-
|
|
276
|
-
|
|
277
|
-
|
|
278
|
-
|
|
279
|
-
|
|
280
|
-
|
|
408
|
+
return scat(
|
|
409
|
+
(self.P00 * other),
|
|
410
|
+
(self.S0 * other),
|
|
411
|
+
(self.S1 * other),
|
|
412
|
+
(self.S2 * other),
|
|
413
|
+
(self.S2L * other),
|
|
414
|
+
self.j1,
|
|
415
|
+
self.j2,
|
|
416
|
+
backend=self.backend,
|
|
417
|
+
)
|
|
418
|
+
|
|
419
|
+
def l1_abs(self, x):
|
|
420
|
+
y = self.get_np(x)
|
|
281
421
|
if self.backend.bk_is_complex(y):
|
|
282
|
-
tmp=y.real*y.real+y.imag*y.imag
|
|
283
|
-
tmp=np.sign(tmp)*np.sqrt(np.fabs(tmp))
|
|
284
|
-
y=tmp
|
|
285
|
-
|
|
286
|
-
return
|
|
287
|
-
|
|
288
|
-
def plot(self,name=None,hold=True,color=
|
|
422
|
+
tmp = y.real * y.real + y.imag * y.imag
|
|
423
|
+
tmp = np.sign(tmp) * np.sqrt(np.fabs(tmp))
|
|
424
|
+
y = tmp
|
|
425
|
+
|
|
426
|
+
return y
|
|
427
|
+
|
|
428
|
+
def plot(self, name=None, hold=True, color="blue", lw=1, legend=True):
|
|
289
429
|
|
|
290
430
|
import matplotlib.pyplot as plt
|
|
291
431
|
|
|
292
|
-
j1,j2=self.get_j_idx()
|
|
293
|
-
|
|
432
|
+
j1, j2 = self.get_j_idx()
|
|
433
|
+
|
|
294
434
|
if name is None:
|
|
295
|
-
name=
|
|
435
|
+
name = ""
|
|
296
436
|
|
|
297
437
|
if hold:
|
|
298
|
-
plt.figure(figsize=(16,8))
|
|
299
|
-
|
|
300
|
-
test=None
|
|
438
|
+
plt.figure(figsize=(16, 8))
|
|
439
|
+
|
|
440
|
+
test = None
|
|
301
441
|
plt.subplot(2, 2, 1)
|
|
302
|
-
tmp=abs(self.get_np(self.S1))
|
|
303
|
-
if len(tmp.shape)==4:
|
|
442
|
+
tmp = abs(self.get_np(self.S1))
|
|
443
|
+
if len(tmp.shape) == 4:
|
|
304
444
|
for k in range(tmp.shape[3]):
|
|
305
445
|
for i1 in range(tmp.shape[0]):
|
|
306
446
|
for i2 in range(tmp.shape[1]):
|
|
307
447
|
if test is None:
|
|
308
|
-
test=1
|
|
309
|
-
plt.plot(
|
|
448
|
+
test = 1
|
|
449
|
+
plt.plot(
|
|
450
|
+
tmp[i1, i2, :, k],
|
|
451
|
+
color=color,
|
|
452
|
+
label=r"%s $S_{1}$" % (name),
|
|
453
|
+
lw=lw,
|
|
454
|
+
)
|
|
310
455
|
else:
|
|
311
|
-
plt.plot(tmp[i1,i2
|
|
456
|
+
plt.plot(tmp[i1, i2, :, k], color=color, lw=lw)
|
|
312
457
|
else:
|
|
313
458
|
for k in range(tmp.shape[2]):
|
|
314
459
|
for i1 in range(tmp.shape[0]):
|
|
315
460
|
if test is None:
|
|
316
|
-
test=1
|
|
317
|
-
plt.plot(
|
|
461
|
+
test = 1
|
|
462
|
+
plt.plot(
|
|
463
|
+
tmp[i1, :, k],
|
|
464
|
+
color=color,
|
|
465
|
+
label=r"%s $S_{1}$" % (name),
|
|
466
|
+
lw=lw,
|
|
467
|
+
)
|
|
318
468
|
else:
|
|
319
|
-
plt.plot(tmp[i1
|
|
320
|
-
plt.yscale(
|
|
321
|
-
plt.ylabel(
|
|
322
|
-
plt.xlabel(r
|
|
469
|
+
plt.plot(tmp[i1, :, k], color=color, lw=lw)
|
|
470
|
+
plt.yscale("log")
|
|
471
|
+
plt.ylabel("S1")
|
|
472
|
+
plt.xlabel(r"$j_{1}$")
|
|
323
473
|
plt.legend()
|
|
324
474
|
|
|
325
|
-
test=None
|
|
475
|
+
test = None
|
|
326
476
|
plt.subplot(2, 2, 2)
|
|
327
|
-
tmp=abs(self.get_np(self.P00))
|
|
328
|
-
if len(tmp.shape)==4:
|
|
477
|
+
tmp = abs(self.get_np(self.P00))
|
|
478
|
+
if len(tmp.shape) == 4:
|
|
329
479
|
for k in range(tmp.shape[3]):
|
|
330
480
|
for i1 in range(tmp.shape[0]):
|
|
331
|
-
for i2 in range(tmp.shape[
|
|
481
|
+
for i2 in range(tmp.shape[1]):
|
|
332
482
|
if test is None:
|
|
333
|
-
test=1
|
|
334
|
-
plt.plot(
|
|
483
|
+
test = 1
|
|
484
|
+
plt.plot(
|
|
485
|
+
tmp[i1, i2, :, k],
|
|
486
|
+
color=color,
|
|
487
|
+
label=r"%s $P_{00}$" % (name),
|
|
488
|
+
lw=lw,
|
|
489
|
+
)
|
|
335
490
|
else:
|
|
336
|
-
plt.plot(tmp[i1,i2
|
|
491
|
+
plt.plot(tmp[i1, i2, :, k], color=color, lw=lw)
|
|
337
492
|
else:
|
|
338
493
|
for k in range(tmp.shape[2]):
|
|
339
494
|
for i1 in range(tmp.shape[0]):
|
|
340
495
|
if test is None:
|
|
341
|
-
test=1
|
|
342
|
-
plt.plot(
|
|
496
|
+
test = 1
|
|
497
|
+
plt.plot(
|
|
498
|
+
tmp[i1, :, k],
|
|
499
|
+
color=color,
|
|
500
|
+
label=r"%s $P_{00}$" % (name),
|
|
501
|
+
lw=lw,
|
|
502
|
+
)
|
|
343
503
|
else:
|
|
344
|
-
plt.plot(tmp[i1
|
|
345
|
-
plt.yscale(
|
|
346
|
-
plt.ylabel(
|
|
347
|
-
plt.xlabel(r
|
|
504
|
+
plt.plot(tmp[i1, :, k], color=color, lw=lw)
|
|
505
|
+
plt.yscale("log")
|
|
506
|
+
plt.ylabel("P00")
|
|
507
|
+
plt.xlabel(r"$j_{1}$")
|
|
348
508
|
plt.legend()
|
|
349
|
-
|
|
350
|
-
ax1=plt.subplot(2, 2, 3)
|
|
509
|
+
|
|
510
|
+
ax1 = plt.subplot(2, 2, 3)
|
|
351
511
|
ax2 = ax1.twiny()
|
|
352
|
-
n=0
|
|
353
|
-
tmp=abs(self.get_np(self.S2))
|
|
354
|
-
lname=r
|
|
355
|
-
ax1.set_ylabel(r
|
|
356
|
-
test=None
|
|
357
|
-
tabx=[]
|
|
358
|
-
tabnx=[]
|
|
359
|
-
tab2x=[]
|
|
360
|
-
tab2nx=[]
|
|
361
|
-
if len(tmp.shape)==5:
|
|
512
|
+
n = 0
|
|
513
|
+
tmp = abs(self.get_np(self.S2))
|
|
514
|
+
lname = r"%s $S_{2}$" % (name)
|
|
515
|
+
ax1.set_ylabel(r"$S_{2}$ [L1 norm]")
|
|
516
|
+
test = None
|
|
517
|
+
tabx = []
|
|
518
|
+
tabnx = []
|
|
519
|
+
tab2x = []
|
|
520
|
+
tab2nx = []
|
|
521
|
+
if len(tmp.shape) == 5:
|
|
362
522
|
for i0 in range(tmp.shape[0]):
|
|
363
523
|
for i1 in range(tmp.shape[1]):
|
|
364
|
-
for i2 in range(j1.max()+1):
|
|
524
|
+
for i2 in range(j1.max() + 1):
|
|
365
525
|
for i3 in range(tmp.shape[3]):
|
|
366
526
|
for i4 in range(tmp.shape[4]):
|
|
367
|
-
if j2[j1==i2].shape[0]==1:
|
|
368
|
-
ax1.plot(
|
|
369
|
-
|
|
527
|
+
if j2[j1 == i2].shape[0] == 1:
|
|
528
|
+
ax1.plot(
|
|
529
|
+
j2[j1 == i2] + n,
|
|
530
|
+
tmp[i0, i1, j1 == i2, i3, i4],
|
|
531
|
+
".",
|
|
532
|
+
color=color,
|
|
533
|
+
lw=lw,
|
|
534
|
+
)
|
|
370
535
|
else:
|
|
371
536
|
if legend and test is None:
|
|
372
|
-
ax1.plot(
|
|
373
|
-
|
|
374
|
-
|
|
375
|
-
|
|
376
|
-
|
|
377
|
-
|
|
378
|
-
|
|
379
|
-
|
|
380
|
-
|
|
381
|
-
|
|
382
|
-
|
|
537
|
+
ax1.plot(
|
|
538
|
+
j2[j1 == i2] + n,
|
|
539
|
+
tmp[i0, i1, j1 == i2, i3, i4],
|
|
540
|
+
color=color,
|
|
541
|
+
label=lname,
|
|
542
|
+
lw=lw,
|
|
543
|
+
)
|
|
544
|
+
test = 1
|
|
545
|
+
ax1.plot(
|
|
546
|
+
j2[j1 == i2] + n,
|
|
547
|
+
tmp[i0, i1, j1 == i2, i3, i4],
|
|
548
|
+
color=color,
|
|
549
|
+
lw=lw,
|
|
550
|
+
)
|
|
551
|
+
tabnx = tabnx + [r"%d" % (k) for k in j2[j1 == i2]]
|
|
552
|
+
tabx = tabx + [k + n for k in j2[j1 == i2]]
|
|
553
|
+
tab2x = tab2x + [(j2[j1 == i2] + n).mean()]
|
|
554
|
+
tab2nx = tab2nx + ["%d" % (i2)]
|
|
555
|
+
ax1.axvline(
|
|
556
|
+
(j2[j1 == i2] + n).max() + 0.5, ls=":", color="gray"
|
|
557
|
+
)
|
|
558
|
+
n = n + j2[j1 == i2].shape[0] - 1
|
|
383
559
|
else:
|
|
384
560
|
for i0 in range(tmp.shape[0]):
|
|
385
|
-
for i2 in range(j1.max()+1):
|
|
561
|
+
for i2 in range(j1.max() + 1):
|
|
386
562
|
for i3 in range(tmp.shape[2]):
|
|
387
563
|
for i4 in range(tmp.shape[3]):
|
|
388
|
-
if j2[j1==i2].shape[0]==1:
|
|
389
|
-
ax1.plot(
|
|
390
|
-
|
|
564
|
+
if j2[j1 == i2].shape[0] == 1:
|
|
565
|
+
ax1.plot(
|
|
566
|
+
j2[j1 == i2] + n,
|
|
567
|
+
tmp[i0, j1 == i2, i3, i4],
|
|
568
|
+
".",
|
|
569
|
+
color=color,
|
|
570
|
+
lw=lw,
|
|
571
|
+
)
|
|
391
572
|
else:
|
|
392
573
|
if legend and test is None:
|
|
393
|
-
ax1.plot(
|
|
394
|
-
|
|
395
|
-
|
|
396
|
-
|
|
397
|
-
|
|
398
|
-
|
|
399
|
-
|
|
400
|
-
|
|
401
|
-
|
|
402
|
-
|
|
403
|
-
|
|
404
|
-
|
|
405
|
-
|
|
574
|
+
ax1.plot(
|
|
575
|
+
j2[j1 == i2] + n,
|
|
576
|
+
tmp[i0, j1 == i2, i3, i4],
|
|
577
|
+
color=color,
|
|
578
|
+
label=lname,
|
|
579
|
+
lw=lw,
|
|
580
|
+
)
|
|
581
|
+
test = 1
|
|
582
|
+
ax1.plot(
|
|
583
|
+
j2[j1 == i2] + n,
|
|
584
|
+
tmp[i0, j1 == i2, i3, i4],
|
|
585
|
+
color=color,
|
|
586
|
+
lw=lw,
|
|
587
|
+
)
|
|
588
|
+
tabnx = tabnx + [r"%d" % (k) for k in j2[j1 == i2]]
|
|
589
|
+
tabx = tabx + [k + n for k in j2[j1 == i2]]
|
|
590
|
+
tab2x = tab2x + [(j2[j1 == i2] + n).mean()]
|
|
591
|
+
tab2nx = tab2nx + ["%d" % (i2)]
|
|
592
|
+
ax1.axvline((j2[j1 == i2] + n).max() + 0.5, ls=":", color="gray")
|
|
593
|
+
n = n + j2[j1 == i2].shape[0] - 1
|
|
594
|
+
plt.yscale("log")
|
|
595
|
+
ax1.set_xlim(0, n + 2)
|
|
406
596
|
ax1.set_xticks(tabx)
|
|
407
|
-
ax1.set_xticklabels(tabnx,fontsize=6)
|
|
408
|
-
ax1.set_xlabel(r"$j_{2}$ ",fontsize=6)
|
|
409
|
-
|
|
597
|
+
ax1.set_xticklabels(tabnx, fontsize=6)
|
|
598
|
+
ax1.set_xlabel(r"$j_{2}$ ", fontsize=6)
|
|
599
|
+
|
|
410
600
|
# Move twinned axis ticks and label from top to bottom
|
|
411
601
|
ax2.xaxis.set_ticks_position("bottom")
|
|
412
602
|
ax2.xaxis.set_label_position("bottom")
|
|
@@ -414,7 +604,7 @@ class scat:
|
|
|
414
604
|
# Offset the twin axis below the host
|
|
415
605
|
ax2.spines["bottom"].set_position(("axes", -0.15))
|
|
416
606
|
|
|
417
|
-
# Turn on the frame for the twin axis, but then hide all
|
|
607
|
+
# Turn on the frame for the twin axis, but then hide all
|
|
418
608
|
# but the bottom spine
|
|
419
609
|
ax2.set_frame_on(True)
|
|
420
610
|
ax2.patch.set_visible(False)
|
|
@@ -422,72 +612,102 @@ class scat:
|
|
|
422
612
|
for sp in ax2.spines.values():
|
|
423
613
|
sp.set_visible(False)
|
|
424
614
|
ax2.spines["bottom"].set_visible(True)
|
|
425
|
-
ax2.set_xlim(0,n+2)
|
|
615
|
+
ax2.set_xlim(0, n + 2)
|
|
426
616
|
ax2.set_xticks(tab2x)
|
|
427
|
-
ax2.set_xticklabels(tab2nx,fontsize=6)
|
|
428
|
-
ax2.set_xlabel(r"$j_{1}$",fontsize=6)
|
|
617
|
+
ax2.set_xticklabels(tab2nx, fontsize=6)
|
|
618
|
+
ax2.set_xlabel(r"$j_{1}$", fontsize=6)
|
|
429
619
|
ax1.legend(frameon=0)
|
|
430
|
-
|
|
431
|
-
ax1=plt.subplot(2, 2, 4)
|
|
620
|
+
|
|
621
|
+
ax1 = plt.subplot(2, 2, 4)
|
|
432
622
|
ax2 = ax1.twiny()
|
|
433
|
-
n=0
|
|
434
|
-
tmp=abs(self.get_np(self.S2L))
|
|
435
|
-
lname=r
|
|
436
|
-
ax1.set_ylabel(r
|
|
437
|
-
test=None
|
|
438
|
-
tabx=[]
|
|
439
|
-
tabnx=[]
|
|
440
|
-
tab2x=[]
|
|
441
|
-
tab2nx=[]
|
|
442
|
-
if len(tmp.shape)==5:
|
|
623
|
+
n = 0
|
|
624
|
+
tmp = abs(self.get_np(self.S2L))
|
|
625
|
+
lname = r"%s $S2_{2}$" % (name)
|
|
626
|
+
ax1.set_ylabel(r"$S_{2}$ [L2 norm]")
|
|
627
|
+
test = None
|
|
628
|
+
tabx = []
|
|
629
|
+
tabnx = []
|
|
630
|
+
tab2x = []
|
|
631
|
+
tab2nx = []
|
|
632
|
+
if len(tmp.shape) == 5:
|
|
443
633
|
for i0 in range(tmp.shape[0]):
|
|
444
634
|
for i1 in range(tmp.shape[1]):
|
|
445
|
-
for i2 in range(j1.max()+1):
|
|
635
|
+
for i2 in range(j1.max() + 1):
|
|
446
636
|
for i3 in range(tmp.shape[3]):
|
|
447
637
|
for i4 in range(tmp.shape[4]):
|
|
448
|
-
if j2[j1==i2].shape[0]==1:
|
|
449
|
-
ax1.plot(
|
|
450
|
-
|
|
638
|
+
if j2[j1 == i2].shape[0] == 1:
|
|
639
|
+
ax1.plot(
|
|
640
|
+
j2[j1 == i2] + n,
|
|
641
|
+
tmp[i0, i1, j1 == i2, i3, i4],
|
|
642
|
+
".",
|
|
643
|
+
color=color,
|
|
644
|
+
lw=lw,
|
|
645
|
+
)
|
|
451
646
|
else:
|
|
452
647
|
if legend and test is None:
|
|
453
|
-
ax1.plot(
|
|
454
|
-
|
|
455
|
-
|
|
456
|
-
|
|
457
|
-
|
|
458
|
-
|
|
459
|
-
|
|
460
|
-
|
|
461
|
-
|
|
462
|
-
|
|
463
|
-
|
|
648
|
+
ax1.plot(
|
|
649
|
+
j2[j1 == i2] + n,
|
|
650
|
+
tmp[i0, i1, j1 == i2, i3, i4],
|
|
651
|
+
color=color,
|
|
652
|
+
label=lname,
|
|
653
|
+
lw=lw,
|
|
654
|
+
)
|
|
655
|
+
test = 1
|
|
656
|
+
ax1.plot(
|
|
657
|
+
j2[j1 == i2] + n,
|
|
658
|
+
tmp[i0, i1, j1 == i2, i3, i4],
|
|
659
|
+
color=color,
|
|
660
|
+
lw=lw,
|
|
661
|
+
)
|
|
662
|
+
tabnx = tabnx + [r"%d" % (k) for k in j2[j1 == i2]]
|
|
663
|
+
tabx = tabx + [k + n for k in j2[j1 == i2]]
|
|
664
|
+
tab2x = tab2x + [(j2[j1 == i2] + n).mean()]
|
|
665
|
+
tab2nx = tab2nx + ["%d" % (i2)]
|
|
666
|
+
ax1.axvline(
|
|
667
|
+
(j2[j1 == i2] + n).max() + 0.5, ls=":", color="gray"
|
|
668
|
+
)
|
|
669
|
+
n = n + j2[j1 == i2].shape[0] - 1
|
|
464
670
|
else:
|
|
465
671
|
for i0 in range(tmp.shape[0]):
|
|
466
|
-
for i2 in range(j1.max()+1):
|
|
672
|
+
for i2 in range(j1.max() + 1):
|
|
467
673
|
for i3 in range(tmp.shape[2]):
|
|
468
674
|
for i4 in range(tmp.shape[3]):
|
|
469
|
-
if j2[j1==i2].shape[0]==1:
|
|
470
|
-
ax1.plot(
|
|
471
|
-
|
|
675
|
+
if j2[j1 == i2].shape[0] == 1:
|
|
676
|
+
ax1.plot(
|
|
677
|
+
j2[j1 == i2] + n,
|
|
678
|
+
tmp[i0, j1 == i2, i3, i4],
|
|
679
|
+
".",
|
|
680
|
+
color=color,
|
|
681
|
+
lw=lw,
|
|
682
|
+
)
|
|
472
683
|
else:
|
|
473
684
|
if legend and test is None:
|
|
474
|
-
ax1.plot(
|
|
475
|
-
|
|
476
|
-
|
|
477
|
-
|
|
478
|
-
|
|
479
|
-
|
|
480
|
-
|
|
481
|
-
|
|
482
|
-
|
|
483
|
-
|
|
484
|
-
|
|
485
|
-
|
|
486
|
-
|
|
685
|
+
ax1.plot(
|
|
686
|
+
j2[j1 == i2] + n,
|
|
687
|
+
tmp[i0, j1 == i2, i3, i4],
|
|
688
|
+
color=color,
|
|
689
|
+
label=lname,
|
|
690
|
+
lw=lw,
|
|
691
|
+
)
|
|
692
|
+
test = 1
|
|
693
|
+
ax1.plot(
|
|
694
|
+
j2[j1 == i2] + n,
|
|
695
|
+
tmp[i0, j1 == i2, i3, i4],
|
|
696
|
+
color=color,
|
|
697
|
+
lw=lw,
|
|
698
|
+
)
|
|
699
|
+
tabnx = tabnx + [r"%d" % (k) for k in j2[j1 == i2]]
|
|
700
|
+
tabx = tabx + [k + n for k in j2[j1 == i2]]
|
|
701
|
+
tab2x = tab2x + [(j2[j1 == i2] + n).mean()]
|
|
702
|
+
tab2nx = tab2nx + ["%d" % (i2)]
|
|
703
|
+
ax1.axvline((j2[j1 == i2] + n).max() + 0.5, ls=":", color="gray")
|
|
704
|
+
n = n + j2[j1 == i2].shape[0] - 1
|
|
705
|
+
plt.yscale("log")
|
|
706
|
+
ax1.set_xlim(-1, n + 3)
|
|
487
707
|
ax1.set_xticks(tabx)
|
|
488
|
-
ax1.set_xticklabels(tabnx,fontsize=6)
|
|
489
|
-
ax1.set_xlabel(r"$j_{2}$",fontsize=6)
|
|
490
|
-
|
|
708
|
+
ax1.set_xticklabels(tabnx, fontsize=6)
|
|
709
|
+
ax1.set_xlabel(r"$j_{2}$", fontsize=6)
|
|
710
|
+
|
|
491
711
|
# Move twinned axis ticks and label from top to bottom
|
|
492
712
|
ax2.xaxis.set_ticks_position("bottom")
|
|
493
713
|
ax2.xaxis.set_label_position("bottom")
|
|
@@ -495,7 +715,7 @@ class scat:
|
|
|
495
715
|
# Offset the twin axis below the host
|
|
496
716
|
ax2.spines["bottom"].set_position(("axes", -0.15))
|
|
497
717
|
|
|
498
|
-
# Turn on the frame for the twin axis, but then hide all
|
|
718
|
+
# Turn on the frame for the twin axis, but then hide all
|
|
499
719
|
# but the bottom spine
|
|
500
720
|
ax2.set_frame_on(True)
|
|
501
721
|
ax2.patch.set_visible(False)
|
|
@@ -503,252 +723,351 @@ class scat:
|
|
|
503
723
|
for sp in ax2.spines.values():
|
|
504
724
|
sp.set_visible(False)
|
|
505
725
|
ax2.spines["bottom"].set_visible(True)
|
|
506
|
-
ax2.set_xlim(0,n+3)
|
|
726
|
+
ax2.set_xlim(0, n + 3)
|
|
507
727
|
ax2.set_xticks(tab2x)
|
|
508
|
-
ax2.set_xticklabels(tab2nx,fontsize=6)
|
|
509
|
-
ax2.set_xlabel(r"$j_{1}$",fontsize=6)
|
|
728
|
+
ax2.set_xticklabels(tab2nx, fontsize=6)
|
|
729
|
+
ax2.set_xlabel(r"$j_{1}$", fontsize=6)
|
|
510
730
|
ax1.legend(frameon=0)
|
|
511
|
-
|
|
512
|
-
def save(self,filename):
|
|
513
|
-
outlist=[
|
|
514
|
-
|
|
515
|
-
|
|
516
|
-
|
|
517
|
-
|
|
518
|
-
|
|
519
|
-
|
|
520
|
-
|
|
521
|
-
|
|
522
|
-
|
|
731
|
+
|
|
732
|
+
def save(self, filename):
|
|
733
|
+
outlist = [
|
|
734
|
+
self.get_S0().numpy(),
|
|
735
|
+
self.get_S1().numpy(),
|
|
736
|
+
self.get_S2().numpy(),
|
|
737
|
+
self.get_S2L().numpy(),
|
|
738
|
+
self.get_P00().numpy(),
|
|
739
|
+
self.j1,
|
|
740
|
+
self.j2,
|
|
741
|
+
]
|
|
742
|
+
|
|
743
|
+
myout = open("%s.pkl" % (filename), "wb")
|
|
744
|
+
pickle.dump(outlist, myout)
|
|
523
745
|
myout.close()
|
|
524
746
|
|
|
525
|
-
|
|
526
|
-
|
|
527
|
-
|
|
528
|
-
|
|
529
|
-
|
|
530
|
-
|
|
531
|
-
|
|
747
|
+
def read(self, filename):
|
|
748
|
+
|
|
749
|
+
outlist = pickle.load(open("%s.pkl" % (filename), "rb"))
|
|
750
|
+
return scat(
|
|
751
|
+
outlist[4],
|
|
752
|
+
outlist[0],
|
|
753
|
+
outlist[1],
|
|
754
|
+
outlist[2],
|
|
755
|
+
outlist[3],
|
|
756
|
+
outlist[5],
|
|
757
|
+
outlist[6],
|
|
758
|
+
backend=bk.foscat_backend("numpy"),
|
|
759
|
+
)
|
|
760
|
+
|
|
761
|
+
def get_np(self, x):
|
|
532
762
|
if isinstance(x, np.ndarray):
|
|
533
763
|
return x
|
|
534
764
|
else:
|
|
535
765
|
return x.numpy()
|
|
536
766
|
|
|
537
767
|
def std(self):
|
|
538
|
-
return np.sqrt(
|
|
539
|
-
|
|
540
|
-
|
|
541
|
-
|
|
542
|
-
|
|
768
|
+
return np.sqrt(
|
|
769
|
+
(
|
|
770
|
+
(abs(self.get_np(self.S0)).std()) ** 2
|
|
771
|
+
+ (abs(self.get_np(self.S1)).std()) ** 2
|
|
772
|
+
+ (abs(self.get_np(self.S2)).std()) ** 2
|
|
773
|
+
+ (abs(self.get_np(self.S2L)).std()) ** 2
|
|
774
|
+
+ (abs(self.get_np(self.P00)).std()) ** 2
|
|
775
|
+
)
|
|
776
|
+
/ 4
|
|
777
|
+
)
|
|
543
778
|
|
|
544
779
|
def mean(self):
|
|
545
|
-
return
|
|
546
|
-
|
|
547
|
-
|
|
548
|
-
|
|
549
|
-
|
|
780
|
+
return (
|
|
781
|
+
abs(
|
|
782
|
+
self.get_np(self.S0).mean()
|
|
783
|
+
+ self.get_np(self.S1).mean()
|
|
784
|
+
+ self.get_np(self.S2).mean()
|
|
785
|
+
+ self.get_np(self.S2L).mean()
|
|
786
|
+
+ self.get_np(self.P00).mean()
|
|
787
|
+
)
|
|
788
|
+
/ 3
|
|
789
|
+
)
|
|
550
790
|
|
|
551
791
|
def sqrt(self):
|
|
552
792
|
|
|
793
|
+
s0 = self.backend.bk_sqrt(self.S0)
|
|
794
|
+
s1 = self.backend.bk_sqrt(self.S1)
|
|
795
|
+
p00 = self.backend.bk_sqrt(self.P00)
|
|
796
|
+
s2 = self.backend.bk_sqrt(self.S2)
|
|
797
|
+
s2L = self.backend.bk_sqrt(self.S2L)
|
|
553
798
|
|
|
554
|
-
s0
|
|
555
|
-
s1 =self.backend.bk_sqrt(self.S1)
|
|
556
|
-
p00=self.backend.bk_sqrt(self.P00)
|
|
557
|
-
s2 =self.backend.bk_sqrt(self.S2)
|
|
558
|
-
s2L=self.backend.bk_sqrt(self.S2L)
|
|
559
|
-
|
|
560
|
-
return scat(p00,s0,s1,s2,s2L,self.j1,self.j2,backend=self.backend)
|
|
561
|
-
|
|
799
|
+
return scat(p00, s0, s1, s2, s2L, self.j1, self.j2, backend=self.backend)
|
|
562
800
|
|
|
563
801
|
def L1(self):
|
|
564
802
|
|
|
803
|
+
s0 = self.backend.bk_L1(self.S0)
|
|
804
|
+
s1 = self.backend.bk_L1(self.S1)
|
|
805
|
+
p00 = self.backend.bk_L1(self.P00)
|
|
806
|
+
s2 = self.backend.bk_L1(self.S2)
|
|
807
|
+
s2L = self.backend.bk_L1(self.S2L)
|
|
808
|
+
|
|
809
|
+
return scat(p00, s0, s1, s2, s2L, self.j1, self.j2, backend=self.backend)
|
|
565
810
|
|
|
566
|
-
s0 =self.backend.bk_L1(self.S0)
|
|
567
|
-
s1 =self.backend.bk_L1(self.S1)
|
|
568
|
-
p00=self.backend.bk_L1(self.P00)
|
|
569
|
-
s2 =self.backend.bk_L1(self.S2)
|
|
570
|
-
s2L=self.backend.bk_L1(self.S2L)
|
|
571
|
-
|
|
572
|
-
return scat(p00,s0,s1,s2,s2L,self.j1,self.j2,backend=self.backend)
|
|
573
|
-
|
|
574
811
|
def square_comp(self):
|
|
575
812
|
|
|
813
|
+
s0 = self.backend.bk_square_comp(self.S0)
|
|
814
|
+
s1 = self.backend.bk_square_comp(self.S1)
|
|
815
|
+
p00 = self.backend.bk_square_comp(self.P00)
|
|
816
|
+
s2 = self.backend.bk_square_comp(self.S2)
|
|
817
|
+
s2L = self.backend.bk_square_comp(self.S2L)
|
|
818
|
+
|
|
819
|
+
return scat(p00, s0, s1, s2, s2L, self.j1, self.j2, backend=self.backend)
|
|
576
820
|
|
|
577
|
-
|
|
578
|
-
|
|
579
|
-
|
|
580
|
-
s2 =self.backend.bk_square_comp(self.S2)
|
|
581
|
-
s2L=self.backend.bk_square_comp(self.S2L)
|
|
582
|
-
|
|
583
|
-
return scat(p00,s0,s1,s2,s2L,self.j1,self.j2,backend=self.backend)
|
|
584
|
-
|
|
585
|
-
def iso_mean(self,repeat=False):
|
|
586
|
-
shape=list(self.S2.shape)
|
|
587
|
-
norient=self.S1.shape[2]
|
|
821
|
+
def iso_mean(self, repeat=False):
|
|
822
|
+
shape = list(self.S2.shape)
|
|
823
|
+
norient = self.S1.shape[2]
|
|
588
824
|
|
|
589
|
-
S1
|
|
825
|
+
S1 = self.backend.bk_reduce_mean(self.S1, 2)
|
|
590
826
|
if repeat:
|
|
591
|
-
S1=self.backend.bk_reshape(
|
|
827
|
+
S1 = self.backend.bk_reshape(
|
|
828
|
+
self.backend.bk_repeat(S1, shape[2], 1), self.S1.shape
|
|
829
|
+
)
|
|
592
830
|
else:
|
|
593
|
-
S1=self.backend.bk_expand_dims(S1
|
|
594
|
-
|
|
831
|
+
S1 = self.backend.bk_expand_dims(S1, -1)
|
|
595
832
|
|
|
596
|
-
P00 = self.backend.bk_reduce_mean(self.P00,2)
|
|
833
|
+
P00 = self.backend.bk_reduce_mean(self.P00, 2)
|
|
597
834
|
if repeat:
|
|
598
|
-
P00=self.backend.bk_reshape(
|
|
835
|
+
P00 = self.backend.bk_reshape(
|
|
836
|
+
self.backend.bk_repeat(P00, shape[2], 1), self.S1.shape
|
|
837
|
+
)
|
|
599
838
|
else:
|
|
600
|
-
P00=self.backend.bk_expand_dims(P00
|
|
839
|
+
P00 = self.backend.bk_expand_dims(P00, -1)
|
|
601
840
|
|
|
602
841
|
if norient not in self.backend._iso_orient:
|
|
603
842
|
self.backend.calc_iso_orient(norient)
|
|
604
|
-
|
|
843
|
+
|
|
605
844
|
if self.backend.bk_is_complex(self.S2):
|
|
606
|
-
lmat
|
|
845
|
+
lmat = self.backend._iso_orient_C[norient]
|
|
607
846
|
lmat_T = self.backend._iso_orient_C_T[norient]
|
|
608
847
|
else:
|
|
609
|
-
lmat
|
|
848
|
+
lmat = self.backend._iso_orient[norient]
|
|
610
849
|
lmat_T = self.backend._iso_orient_T[norient]
|
|
611
|
-
|
|
612
|
-
S2=self.backend.bk_reshape(
|
|
613
|
-
self.backend.backend.matmul(
|
|
614
|
-
|
|
615
|
-
|
|
616
|
-
|
|
617
|
-
|
|
618
|
-
|
|
850
|
+
|
|
851
|
+
S2 = self.backend.bk_reshape(
|
|
852
|
+
self.backend.backend.matmul(
|
|
853
|
+
self.backend.bk_reshape(
|
|
854
|
+
self.S2, [shape[0], shape[1], norient * norient]
|
|
855
|
+
),
|
|
856
|
+
lmat,
|
|
857
|
+
),
|
|
858
|
+
[shape[0], shape[1], norient],
|
|
859
|
+
)
|
|
860
|
+
S2L = self.backend.bk_reshape(
|
|
861
|
+
self.backend.backend.matmul(
|
|
862
|
+
self.backend.bk_reshape(
|
|
863
|
+
self.S2L, [shape[0], shape[1], norient * norient]
|
|
864
|
+
),
|
|
865
|
+
lmat,
|
|
866
|
+
),
|
|
867
|
+
[shape[0], shape[1], norient],
|
|
868
|
+
)
|
|
869
|
+
|
|
619
870
|
if repeat:
|
|
620
|
-
|
|
621
|
-
S2=self.backend.bk_reshape(
|
|
622
|
-
self.backend.backend.matmul(
|
|
623
|
-
|
|
624
|
-
|
|
625
|
-
self.
|
|
626
|
-
|
|
871
|
+
|
|
872
|
+
S2 = self.backend.bk_reshape(
|
|
873
|
+
self.backend.backend.matmul(
|
|
874
|
+
self.backend.bk_reshape(S2, [shape[0] * shape[1], norient]), lmat_T
|
|
875
|
+
),
|
|
876
|
+
self.S2.shape,
|
|
877
|
+
)
|
|
878
|
+
S2L = self.backend.bk_reshape(
|
|
879
|
+
self.backend.backend.matmul(
|
|
880
|
+
self.backend.bk_reshape(S2L, [shape[0] * shape[1], norient]), lmat_T
|
|
881
|
+
),
|
|
882
|
+
self.S2.shape,
|
|
883
|
+
)
|
|
627
884
|
else:
|
|
628
|
-
S2=self.backend.bk_expand_dims(S2
|
|
629
|
-
S2L=self.backend.bk_expand_dims(S2L
|
|
885
|
+
S2 = self.backend.bk_expand_dims(S2, -1)
|
|
886
|
+
S2L = self.backend.bk_expand_dims(S2L, -1)
|
|
630
887
|
|
|
631
|
-
return scat(P00,self.S0,S1,S2,S2L,self.j1,self.j2,backend=self.backend)
|
|
888
|
+
return scat(P00, self.S0, S1, S2, S2L, self.j1, self.j2, backend=self.backend)
|
|
632
889
|
|
|
633
|
-
|
|
634
|
-
|
|
635
|
-
|
|
636
|
-
norient=self.S1.shape[2]
|
|
890
|
+
def fft_ang(self, nharm=1, imaginary=False):
|
|
891
|
+
shape = list(self.S2.shape)
|
|
892
|
+
norient = self.S1.shape[2]
|
|
637
893
|
|
|
638
|
-
nout=1+nharm
|
|
894
|
+
nout = 1 + nharm
|
|
639
895
|
if imaginary:
|
|
640
|
-
nout=1+nharm*2
|
|
641
|
-
|
|
642
|
-
if (norient,nharm) not in self.backend._fft_1_orient:
|
|
643
|
-
self.backend.calc_fft_orient(norient,nharm,imaginary)
|
|
644
|
-
|
|
896
|
+
nout = 1 + nharm * 2
|
|
897
|
+
|
|
898
|
+
if (norient, nharm) not in self.backend._fft_1_orient:
|
|
899
|
+
self.backend.calc_fft_orient(norient, nharm, imaginary)
|
|
900
|
+
|
|
645
901
|
if self.backend.bk_is_complex(self.S1):
|
|
646
|
-
lmat
|
|
902
|
+
lmat = self.backend._fft_1_orient_C[(norient, nharm, imaginary)]
|
|
647
903
|
else:
|
|
648
|
-
lmat
|
|
649
|
-
|
|
650
|
-
S1=self.backend.bk_reshape(
|
|
651
|
-
self.backend.backend.matmul(
|
|
652
|
-
|
|
653
|
-
|
|
654
|
-
|
|
655
|
-
|
|
656
|
-
|
|
657
|
-
|
|
658
|
-
|
|
904
|
+
lmat = self.backend._fft_1_orient[(norient, nharm, imaginary)]
|
|
905
|
+
|
|
906
|
+
S1 = self.backend.bk_reshape(
|
|
907
|
+
self.backend.backend.matmul(
|
|
908
|
+
self.backend.bk_reshape(
|
|
909
|
+
self.S1, [self.S1.shape[0], self.S1.shape[1], norient]
|
|
910
|
+
),
|
|
911
|
+
lmat,
|
|
912
|
+
),
|
|
913
|
+
[self.S1.shape[0], self.S1.shape[1], nout],
|
|
914
|
+
)
|
|
915
|
+
|
|
916
|
+
P00 = self.backend.bk_reshape(
|
|
917
|
+
self.backend.backend.matmul(
|
|
918
|
+
self.backend.bk_reshape(
|
|
919
|
+
self.P00, [self.S1.shape[0], self.S1.shape[1], norient]
|
|
920
|
+
),
|
|
921
|
+
lmat,
|
|
922
|
+
),
|
|
923
|
+
[self.S1.shape[0], self.S1.shape[1], nout],
|
|
924
|
+
)
|
|
925
|
+
|
|
659
926
|
if self.backend.bk_is_complex(self.S2):
|
|
660
|
-
lmat
|
|
927
|
+
lmat = self.backend._fft_2_orient_C[(norient, nharm, imaginary)]
|
|
661
928
|
else:
|
|
662
|
-
lmat
|
|
663
|
-
|
|
664
|
-
S2=self.backend.bk_reshape(
|
|
665
|
-
self.backend.backend.matmul(
|
|
666
|
-
|
|
667
|
-
|
|
668
|
-
|
|
669
|
-
|
|
670
|
-
|
|
671
|
-
|
|
672
|
-
|
|
673
|
-
|
|
674
|
-
|
|
675
|
-
|
|
676
|
-
|
|
929
|
+
lmat = self.backend._fft_2_orient[(norient, nharm, imaginary)]
|
|
930
|
+
|
|
931
|
+
S2 = self.backend.bk_reshape(
|
|
932
|
+
self.backend.backend.matmul(
|
|
933
|
+
self.backend.bk_reshape(
|
|
934
|
+
self.S2, [shape[0], shape[1], norient * norient]
|
|
935
|
+
),
|
|
936
|
+
lmat,
|
|
937
|
+
),
|
|
938
|
+
[shape[0], shape[1], nout, nout],
|
|
939
|
+
)
|
|
940
|
+
S2L = self.backend.bk_reshape(
|
|
941
|
+
self.backend.backend.matmul(
|
|
942
|
+
self.backend.bk_reshape(
|
|
943
|
+
self.S2L, [shape[0], shape[1], norient * norient]
|
|
944
|
+
),
|
|
945
|
+
lmat,
|
|
946
|
+
),
|
|
947
|
+
[shape[0], shape[1], nout, nout],
|
|
948
|
+
)
|
|
949
|
+
|
|
950
|
+
return scat(P00, self.S0, S1, S2, S2L, self.j1, self.j2, backend=self.backend)
|
|
951
|
+
|
|
952
|
+
def iso_std(self, repeat=False):
|
|
953
|
+
|
|
954
|
+
val = (self - self.iso_mean(repeat=True)).square_comp()
|
|
677
955
|
return (val.iso_mean(repeat=repeat)).L1()
|
|
678
956
|
|
|
679
957
|
# ---------------------------------------------−---------
|
|
680
|
-
def cleanval(self,x):
|
|
681
|
-
x=x.numpy()
|
|
682
|
-
x[np.isfinite(x)
|
|
958
|
+
def cleanval(self, x):
|
|
959
|
+
x = x.numpy()
|
|
960
|
+
x[~np.isfinite(x)] = np.median(x[np.isfinite(x)])
|
|
683
961
|
return x
|
|
684
962
|
|
|
685
963
|
def filter_inf(self):
|
|
686
|
-
S1
|
|
687
|
-
S0
|
|
964
|
+
S1 = self.cleanval(self.S1)
|
|
965
|
+
S0 = self.cleanval(self.S0)
|
|
688
966
|
P00 = self.cleanval(self.P00)
|
|
689
|
-
S2
|
|
967
|
+
S2 = self.cleanval(self.S2)
|
|
690
968
|
S2L = self.cleanval(self.S2L)
|
|
691
969
|
|
|
692
|
-
return scat(P00,S0,S1,S2,S2L,self.j1,self.j2,backend=self.backend)
|
|
970
|
+
return scat(P00, S0, S1, S2, S2L, self.j1, self.j2, backend=self.backend)
|
|
693
971
|
|
|
694
972
|
# ---------------------------------------------−---------
|
|
695
|
-
def interp(
|
|
696
|
-
|
|
697
|
-
|
|
698
|
-
|
|
699
|
-
|
|
700
|
-
|
|
701
|
-
|
|
702
|
-
|
|
703
|
-
|
|
704
|
-
|
|
973
|
+
def interp(
|
|
974
|
+
self, nscale, extend=False, constant=False, threshold=1e30, use_mask=False
|
|
975
|
+
):
|
|
976
|
+
|
|
977
|
+
if nscale + 2 > self.S1.shape[1]:
|
|
978
|
+
print(
|
|
979
|
+
"Can not *interp* %d with a statistic described over %d"
|
|
980
|
+
% (nscale, self.S1.shape[1])
|
|
981
|
+
)
|
|
982
|
+
return scat(
|
|
983
|
+
self.P00,
|
|
984
|
+
self.S0,
|
|
985
|
+
self.S1,
|
|
986
|
+
self.S2,
|
|
987
|
+
self.S2L,
|
|
988
|
+
self.j1,
|
|
989
|
+
self.j2,
|
|
990
|
+
backend=self.backend,
|
|
991
|
+
)
|
|
992
|
+
if isinstance(self.S1, np.ndarray):
|
|
993
|
+
s1 = self.S1
|
|
994
|
+
p0 = self.P00
|
|
995
|
+
s2 = self.S2
|
|
996
|
+
s2l = self.S2L
|
|
705
997
|
else:
|
|
706
|
-
s1=self.S1.numpy()
|
|
707
|
-
p0=self.P00.numpy()
|
|
708
|
-
s2=self.S2.numpy()
|
|
709
|
-
s2l=self.S2L.numpy()
|
|
710
|
-
|
|
711
|
-
print(s1.sum(),p0.sum(),s2.sum(),s2l.sum())
|
|
712
|
-
|
|
713
|
-
if isinstance(threshold,scat):
|
|
714
|
-
if isinstance(threshold.S1,np.ndarray):
|
|
715
|
-
s1th=threshold.S1
|
|
716
|
-
p0th=threshold.P00
|
|
717
|
-
s2th=threshold.S2
|
|
718
|
-
s2lth=threshold.S2L
|
|
998
|
+
s1 = self.S1.numpy()
|
|
999
|
+
p0 = self.P00.numpy()
|
|
1000
|
+
s2 = self.S2.numpy()
|
|
1001
|
+
s2l = self.S2L.numpy()
|
|
1002
|
+
|
|
1003
|
+
print(s1.sum(), p0.sum(), s2.sum(), s2l.sum())
|
|
1004
|
+
|
|
1005
|
+
if isinstance(threshold, scat):
|
|
1006
|
+
if isinstance(threshold.S1, np.ndarray):
|
|
1007
|
+
s1th = threshold.S1
|
|
1008
|
+
p0th = threshold.P00
|
|
1009
|
+
s2th = threshold.S2
|
|
1010
|
+
s2lth = threshold.S2L
|
|
719
1011
|
else:
|
|
720
|
-
s1th=threshold.S1.numpy()
|
|
721
|
-
p0th=threshold.P00.numpy()
|
|
722
|
-
s2th=threshold.S2.numpy()
|
|
723
|
-
s2lth=threshold.S2L.numpy()
|
|
1012
|
+
s1th = threshold.S1.numpy()
|
|
1013
|
+
p0th = threshold.P00.numpy()
|
|
1014
|
+
s2th = threshold.S2.numpy()
|
|
1015
|
+
s2lth = threshold.S2L.numpy()
|
|
724
1016
|
else:
|
|
725
|
-
s1th=threshold+0*s1
|
|
726
|
-
p0th=threshold+0*p0
|
|
727
|
-
s2th=threshold+0*s2
|
|
728
|
-
s2lth=threshold+0*s2l
|
|
1017
|
+
s1th = threshold + 0 * s1
|
|
1018
|
+
p0th = threshold + 0 * p0
|
|
1019
|
+
s2th = threshold + 0 * s2
|
|
1020
|
+
s2lth = threshold + 0 * s2l
|
|
729
1021
|
|
|
730
1022
|
for k in range(nscale):
|
|
731
1023
|
if constant:
|
|
732
|
-
s1[:,nscale-1-k
|
|
733
|
-
p0[:,nscale-1-k
|
|
1024
|
+
s1[:, nscale - 1 - k, :] = s1[:, nscale - k, :]
|
|
1025
|
+
p0[:, nscale - 1 - k, :] = p0[:, nscale - k, :]
|
|
734
1026
|
else:
|
|
735
|
-
idx=np.where(
|
|
736
|
-
|
|
737
|
-
|
|
738
|
-
|
|
739
|
-
|
|
740
|
-
|
|
741
|
-
|
|
742
|
-
|
|
743
|
-
|
|
744
|
-
|
|
745
|
-
|
|
746
|
-
|
|
747
|
-
|
|
748
|
-
|
|
749
|
-
|
|
750
|
-
|
|
751
|
-
|
|
1027
|
+
idx = np.where(
|
|
1028
|
+
(s1[:, nscale + 1 - k, :] > 0)
|
|
1029
|
+
* (s1[:, nscale + 2 - k, :] > 0)
|
|
1030
|
+
* (s1[:, nscale - k, :] < s1th[:, nscale - k, :])
|
|
1031
|
+
)
|
|
1032
|
+
if len(idx[0]) > 0:
|
|
1033
|
+
s1[idx[0], nscale - 1 - k, idx[1]] = np.exp(
|
|
1034
|
+
3 * np.log(s1[idx[0], nscale + 1 - k, idx[1]])
|
|
1035
|
+
- 2 * np.log(s1[idx[0], nscale + 2 - k, idx[1]])
|
|
1036
|
+
)
|
|
1037
|
+
idx = np.where(
|
|
1038
|
+
(s1[:, nscale - k, :] > 0)
|
|
1039
|
+
* (s1[:, nscale + 1 - k, :] > 0)
|
|
1040
|
+
* (s1[:, nscale - 1 - k, :] < s1th[:, nscale - 1 - k, :])
|
|
1041
|
+
)
|
|
1042
|
+
if len(idx[0]) > 0:
|
|
1043
|
+
s1[idx[0], nscale - 1 - k, idx[1]] = np.exp(
|
|
1044
|
+
2 * np.log(s1[idx[0], nscale - k, idx[1]])
|
|
1045
|
+
- np.log(s1[idx[0], nscale + 1 - k, idx[1]])
|
|
1046
|
+
)
|
|
1047
|
+
|
|
1048
|
+
idx = np.where(
|
|
1049
|
+
(p0[:, nscale + 1 - k, :] > 0)
|
|
1050
|
+
* (p0[:, nscale + 2 - k, :] > 0)
|
|
1051
|
+
* (p0[:, nscale - k, :] < p0th[:, nscale - k, :])
|
|
1052
|
+
)
|
|
1053
|
+
if len(idx[0]) > 0:
|
|
1054
|
+
p0[idx[0], nscale - 1 - k, idx[1]] = np.exp(
|
|
1055
|
+
3 * np.log(p0[idx[0], nscale + 1 - k, idx[1]])
|
|
1056
|
+
- 2 * np.log(p0[idx[0], nscale + 2 - k, idx[1]])
|
|
1057
|
+
)
|
|
1058
|
+
|
|
1059
|
+
idx = np.where(
|
|
1060
|
+
(p0[:, nscale - k, :] > 0)
|
|
1061
|
+
* (p0[:, nscale + 1 - k, :] > 0)
|
|
1062
|
+
* (p0[:, nscale - 1 - k, :] < p0th[:, nscale - 1 - k, :])
|
|
1063
|
+
)
|
|
1064
|
+
if len(idx[0]) > 0:
|
|
1065
|
+
p0[idx[0], nscale - 1 - k, idx[1]] = np.exp(
|
|
1066
|
+
2 * np.log(p0[idx[0], nscale - k, idx[1]])
|
|
1067
|
+
- np.log(p0[idx[0], nscale + 1 - k, idx[1]])
|
|
1068
|
+
)
|
|
1069
|
+
|
|
1070
|
+
j1, j2 = self.get_j_idx()
|
|
752
1071
|
|
|
753
1072
|
for k in range(nscale):
|
|
754
1073
|
|
|
@@ -761,646 +1080,951 @@ class scat:
|
|
|
761
1080
|
s2l[:,i0]=np.exp(2*np.log(s2l[:,i1])-np.log(s2l[:,i2]))
|
|
762
1081
|
"""
|
|
763
1082
|
|
|
764
|
-
for
|
|
765
|
-
i0=np.where(
|
|
766
|
-
|
|
767
|
-
|
|
768
|
-
|
|
769
|
-
|
|
1083
|
+
for l_scale in range(nscale - k):
|
|
1084
|
+
i0 = np.where(
|
|
1085
|
+
(j1 == nscale - 1 - k - l_scale) * (j2 == nscale - 1 - k)
|
|
1086
|
+
)[0]
|
|
1087
|
+
i1 = np.where((j1 == nscale - 1 - k - l_scale) * (j2 == nscale - k))[0]
|
|
1088
|
+
i2 = np.where(
|
|
1089
|
+
(j1 == nscale - 1 - k - l_scale) * (j2 == nscale + 1 - k)
|
|
1090
|
+
)[0]
|
|
1091
|
+
i3 = np.where(
|
|
1092
|
+
(j1 == nscale - 1 - k - l_scale) * (j2 == nscale + 2 - k)
|
|
1093
|
+
)[0]
|
|
1094
|
+
|
|
770
1095
|
if constant:
|
|
771
|
-
s2[:,i0]=s2[:,i1]
|
|
772
|
-
s2l[:,i0]=s2l[:,i1]
|
|
1096
|
+
s2[:, i0] = s2[:, i1]
|
|
1097
|
+
s2l[:, i0] = s2l[:, i1]
|
|
773
1098
|
else:
|
|
774
|
-
idx=np.where(
|
|
775
|
-
|
|
776
|
-
|
|
777
|
-
|
|
778
|
-
|
|
779
|
-
|
|
780
|
-
|
|
781
|
-
|
|
782
|
-
|
|
783
|
-
|
|
784
|
-
|
|
785
|
-
|
|
786
|
-
idx
|
|
787
|
-
|
|
788
|
-
|
|
789
|
-
|
|
1099
|
+
idx = np.where(
|
|
1100
|
+
(s2[:, i2] > 0) * (s2[:, i3] > 0) * (s2[:, i2] < s2th[:, i2])
|
|
1101
|
+
)
|
|
1102
|
+
if len(idx[0]) > 0:
|
|
1103
|
+
s2[idx[0], i0, idx[1], idx[2]] = np.exp(
|
|
1104
|
+
3 * np.log(s2[idx[0], i2, idx[1], idx[2]])
|
|
1105
|
+
- 2 * np.log(s2[idx[0], i3, idx[1], idx[2]])
|
|
1106
|
+
)
|
|
1107
|
+
|
|
1108
|
+
idx = np.where(
|
|
1109
|
+
(s2[:, i1] > 0) * (s2[:, i2] > 0) * (s2[:, i1] < s2th[:, i1])
|
|
1110
|
+
)
|
|
1111
|
+
if len(idx[0]) > 0:
|
|
1112
|
+
s2[idx[0], i0, idx[1], idx[2]] = np.exp(
|
|
1113
|
+
2 * np.log(s2[idx[0], i1, idx[1], idx[2]])
|
|
1114
|
+
- np.log(s2[idx[0], i2, idx[1], idx[2]])
|
|
1115
|
+
)
|
|
1116
|
+
|
|
1117
|
+
idx = np.where(
|
|
1118
|
+
(s2l[:, i2] > 0)
|
|
1119
|
+
* (s2l[:, i3] > 0)
|
|
1120
|
+
* (s2l[:, i2] < s2lth[:, i2])
|
|
1121
|
+
)
|
|
1122
|
+
if len(idx[0]) > 0:
|
|
1123
|
+
s2l[idx[0], i0, idx[1], idx[2]] = np.exp(
|
|
1124
|
+
3 * np.log(s2l[idx[0], i2, idx[1], idx[2]])
|
|
1125
|
+
- 2 * np.log(s2l[idx[0], i3, idx[1], idx[2]])
|
|
1126
|
+
)
|
|
1127
|
+
|
|
1128
|
+
idx = np.where(
|
|
1129
|
+
(s2l[:, i1] > 0)
|
|
1130
|
+
* (s2l[:, i2] > 0)
|
|
1131
|
+
* (s2l[:, i1] < s2lth[:, i1])
|
|
1132
|
+
)
|
|
1133
|
+
if len(idx[0]) > 0:
|
|
1134
|
+
s2l[idx[0], i0, idx[1], idx[2]] = np.exp(
|
|
1135
|
+
2 * np.log(s2l[idx[0], i1, idx[1], idx[2]])
|
|
1136
|
+
- np.log(s2l[idx[0], i2, idx[1], idx[2]])
|
|
1137
|
+
)
|
|
1138
|
+
|
|
790
1139
|
if extend:
|
|
791
1140
|
for k in range(nscale):
|
|
792
|
-
for
|
|
793
|
-
i0=np.where(
|
|
794
|
-
|
|
795
|
-
|
|
796
|
-
|
|
1141
|
+
for l_scale in range(1, nscale):
|
|
1142
|
+
i0 = np.where(
|
|
1143
|
+
(j1 == 2 * nscale - 1 - k)
|
|
1144
|
+
* (j2 == 2 * nscale - 1 - k - l_scale)
|
|
1145
|
+
)[0]
|
|
1146
|
+
i1 = np.where(
|
|
1147
|
+
(j1 == 2 * nscale - 1 - k) * (j2 == 2 * nscale - k - l_scale)
|
|
1148
|
+
)[0]
|
|
1149
|
+
i2 = np.where(
|
|
1150
|
+
(j1 == 2 * nscale - 1 - k)
|
|
1151
|
+
* (j2 == 2 * nscale + 1 - k - l_scale)
|
|
1152
|
+
)[0]
|
|
1153
|
+
i3 = np.where(
|
|
1154
|
+
(j1 == 2 * nscale - 1 - k)
|
|
1155
|
+
* (j2 == 2 * nscale + 2 - k - l_scale)
|
|
1156
|
+
)[0]
|
|
797
1157
|
if constant:
|
|
798
|
-
s2[:,i0]=s2[:,i1]
|
|
799
|
-
s2l[:,i0]=s2l[:,i1]
|
|
1158
|
+
s2[:, i0] = s2[:, i1]
|
|
1159
|
+
s2l[:, i0] = s2l[:, i1]
|
|
800
1160
|
else:
|
|
801
|
-
idx=np.where(
|
|
802
|
-
|
|
803
|
-
|
|
804
|
-
|
|
805
|
-
|
|
806
|
-
|
|
807
|
-
|
|
808
|
-
|
|
809
|
-
|
|
810
|
-
|
|
811
|
-
idx=np.where(
|
|
812
|
-
|
|
813
|
-
|
|
814
|
-
|
|
815
|
-
|
|
816
|
-
|
|
817
|
-
|
|
818
|
-
|
|
819
|
-
|
|
820
|
-
|
|
821
|
-
|
|
822
|
-
|
|
823
|
-
|
|
824
|
-
|
|
1161
|
+
idx = np.where(
|
|
1162
|
+
(s2[:, i2] > 0)
|
|
1163
|
+
* (s2[:, i3] > 0)
|
|
1164
|
+
* (s2[:, i2] < s2th[:, i2])
|
|
1165
|
+
)
|
|
1166
|
+
if len(idx[0]) > 0:
|
|
1167
|
+
s2[idx[0], i0, idx[1], idx[2]] = np.exp(
|
|
1168
|
+
3 * np.log(s2[idx[0], i2, idx[1], idx[2]])
|
|
1169
|
+
- 2 * np.log(s2[idx[0], i3, idx[1], idx[2]])
|
|
1170
|
+
)
|
|
1171
|
+
idx = np.where(
|
|
1172
|
+
(s2[:, i1] > 0)
|
|
1173
|
+
* (s2[:, i2] > 0)
|
|
1174
|
+
* (s2[:, i1] < s2th[:, i1])
|
|
1175
|
+
)
|
|
1176
|
+
if len(idx[0]) > 0:
|
|
1177
|
+
s2[idx[0], i0, idx[1], idx[2]] = np.exp(
|
|
1178
|
+
2 * np.log(s2[idx[0], i1, idx[1], idx[2]])
|
|
1179
|
+
- np.log(s2[idx[0], i2, idx[1], idx[2]])
|
|
1180
|
+
)
|
|
1181
|
+
|
|
1182
|
+
idx = np.where(
|
|
1183
|
+
(s2l[:, i2] > 0)
|
|
1184
|
+
* (s2l[:, i3] > 0)
|
|
1185
|
+
* (s2l[:, i2] < s2lth[:, i2])
|
|
1186
|
+
)
|
|
1187
|
+
if len(idx[0]) > 0:
|
|
1188
|
+
s2l[idx[0], i0, idx[1], idx[2]] = np.exp(
|
|
1189
|
+
3 * np.log(s2l[idx[0], i2, idx[1], idx[2]])
|
|
1190
|
+
- 2 * np.log(s2l[idx[0], i3, idx[1], idx[2]])
|
|
1191
|
+
)
|
|
1192
|
+
idx = np.where(
|
|
1193
|
+
(s2l[:, i1] > 0)
|
|
1194
|
+
* (s2l[:, i2] > 0)
|
|
1195
|
+
* (s2l[:, i1] < s2lth[:, i1])
|
|
1196
|
+
)
|
|
1197
|
+
if len(idx[0]) > 0:
|
|
1198
|
+
s2l[idx[0], i0, idx[1], idx[2]] = np.exp(
|
|
1199
|
+
2 * np.log(s2l[idx[0], i1, idx[1], idx[2]])
|
|
1200
|
+
- np.log(s2l[idx[0], i2, idx[1], idx[2]])
|
|
1201
|
+
)
|
|
1202
|
+
|
|
1203
|
+
s1[np.isnan(s1)] = 0.0
|
|
1204
|
+
p0[np.isnan(p0)] = 0.0
|
|
1205
|
+
s2[np.isnan(s2)] = 0.0
|
|
1206
|
+
s2l[np.isnan(s2l)] = 0.0
|
|
1207
|
+
print(s1.sum(), p0.sum(), s2.sum(), s2l.sum())
|
|
1208
|
+
|
|
1209
|
+
return scat(
|
|
1210
|
+
self.backend.constant(p0),
|
|
1211
|
+
self.S0,
|
|
1212
|
+
self.backend.constant(s1),
|
|
1213
|
+
self.backend.constant(s2),
|
|
1214
|
+
self.backend.constant(s2l),
|
|
1215
|
+
self.j1,
|
|
1216
|
+
self.j2,
|
|
1217
|
+
backend=self.backend,
|
|
1218
|
+
)
|
|
825
1219
|
|
|
826
1220
|
# ---------------------------------------------−---------
|
|
827
1221
|
def flatten(self):
|
|
828
|
-
if isinstance(self.S1,np.ndarray):
|
|
829
|
-
return np.concatenate(
|
|
830
|
-
|
|
831
|
-
|
|
832
|
-
|
|
833
|
-
|
|
1222
|
+
if isinstance(self.S1, np.ndarray):
|
|
1223
|
+
return np.concatenate(
|
|
1224
|
+
[
|
|
1225
|
+
self.S0.flatten(),
|
|
1226
|
+
self.S1.flatten(),
|
|
1227
|
+
self.P00.flatten(),
|
|
1228
|
+
self.S2.flatten(),
|
|
1229
|
+
self.S2L.flatten(),
|
|
1230
|
+
],
|
|
1231
|
+
0,
|
|
1232
|
+
)
|
|
834
1233
|
else:
|
|
835
|
-
return self.backend.bk_concat(
|
|
836
|
-
|
|
837
|
-
|
|
838
|
-
|
|
839
|
-
|
|
1234
|
+
return self.backend.bk_concat(
|
|
1235
|
+
[
|
|
1236
|
+
self.backend.bk_flattenR(self.S0),
|
|
1237
|
+
self.backend.bk_flattenR(self.S1),
|
|
1238
|
+
self.backend.bk_flattenR(self.P00),
|
|
1239
|
+
self.backend.bk_flattenR(self.S2),
|
|
1240
|
+
self.backend.bk_flattenR(self.S2),
|
|
1241
|
+
],
|
|
1242
|
+
axis=0,
|
|
1243
|
+
)
|
|
840
1244
|
|
|
841
1245
|
# ---------------------------------------------−---------
|
|
842
1246
|
def flattenMask(self):
|
|
843
|
-
if isinstance(self.S1,np.ndarray):
|
|
844
|
-
tmp=np.expand_dims(
|
|
845
|
-
|
|
846
|
-
|
|
847
|
-
|
|
848
|
-
|
|
849
|
-
|
|
850
|
-
|
|
851
|
-
|
|
852
|
-
|
|
853
|
-
|
|
854
|
-
|
|
855
|
-
|
|
1247
|
+
if isinstance(self.S1, np.ndarray):
|
|
1248
|
+
tmp = np.expand_dims(
|
|
1249
|
+
np.concatenate(
|
|
1250
|
+
[
|
|
1251
|
+
self.S1[0].flatten(),
|
|
1252
|
+
self.P00[0].flatten(),
|
|
1253
|
+
self.S2[0].flatten(),
|
|
1254
|
+
self.S2L[0].flatten(),
|
|
1255
|
+
],
|
|
1256
|
+
0,
|
|
1257
|
+
),
|
|
1258
|
+
0,
|
|
1259
|
+
)
|
|
1260
|
+
for k in range(1, self.P00.shape[0]):
|
|
1261
|
+
tmp = np.concatenate(
|
|
1262
|
+
[
|
|
1263
|
+
tmp,
|
|
1264
|
+
np.expand_dims(
|
|
1265
|
+
np.concatenate(
|
|
1266
|
+
[
|
|
1267
|
+
self.S1[k].flatten(),
|
|
1268
|
+
self.P00[k].flatten(),
|
|
1269
|
+
self.S2[k].flatten(),
|
|
1270
|
+
self.S2L[k].flatten(),
|
|
1271
|
+
],
|
|
1272
|
+
0,
|
|
1273
|
+
),
|
|
1274
|
+
0,
|
|
1275
|
+
),
|
|
1276
|
+
],
|
|
1277
|
+
0,
|
|
1278
|
+
)
|
|
1279
|
+
|
|
1280
|
+
return np.concatenate([tmp, np.expand_dims(self.S0, 1)], 1)
|
|
856
1281
|
else:
|
|
857
|
-
tmp=self.backend.bk_expand_dims(
|
|
858
|
-
|
|
859
|
-
|
|
860
|
-
|
|
861
|
-
|
|
862
|
-
|
|
863
|
-
|
|
864
|
-
|
|
865
|
-
|
|
866
|
-
|
|
867
|
-
|
|
868
|
-
|
|
869
|
-
|
|
1282
|
+
tmp = self.backend.bk_expand_dims(
|
|
1283
|
+
self.backend.bk_concat(
|
|
1284
|
+
[
|
|
1285
|
+
self.backend.bk_flattenR(self.S1[0]),
|
|
1286
|
+
self.backend.bk_flattenR(self.P00[0]),
|
|
1287
|
+
self.backend.bk_flattenR(self.S2[0]),
|
|
1288
|
+
self.backend.bk_flattenR(self.S2[0]),
|
|
1289
|
+
],
|
|
1290
|
+
axis=0,
|
|
1291
|
+
),
|
|
1292
|
+
0,
|
|
1293
|
+
)
|
|
1294
|
+
for k in range(1, self.P00.shape[0]):
|
|
1295
|
+
ltmp = self.backend.bk_expand_dims(
|
|
1296
|
+
self.backend.bk_concat(
|
|
1297
|
+
[
|
|
1298
|
+
self.backend.bk_flattenR(self.S1[k]),
|
|
1299
|
+
self.backend.bk_flattenR(self.P00[k]),
|
|
1300
|
+
self.backend.bk_flattenR(self.S2[k]),
|
|
1301
|
+
self.backend.bk_flattenR(self.S2[k]),
|
|
1302
|
+
],
|
|
1303
|
+
axis=0,
|
|
1304
|
+
),
|
|
1305
|
+
0,
|
|
1306
|
+
)
|
|
1307
|
+
tmp = self.backend.bk_concat([tmp, ltmp], 0)
|
|
1308
|
+
|
|
1309
|
+
return self.backend.bk_concat(
|
|
1310
|
+
[tmp, self.backend.bk_expand_dims(self.S0, 1)], 1
|
|
1311
|
+
)
|
|
1312
|
+
|
|
870
1313
|
# ---------------------------------------------−---------
|
|
871
|
-
def model(self,i__y,add=0,dx=3,dell=2,weigth=None,inverse=False):
|
|
1314
|
+
def model(self, i__y, add=0, dx=3, dell=2, weigth=None, inverse=False):
|
|
872
1315
|
|
|
873
|
-
if i__y.shape[0]<dx+1:
|
|
874
|
-
l__dx=i__y.shape[0]-1
|
|
1316
|
+
if i__y.shape[0] < dx + 1:
|
|
1317
|
+
l__dx = i__y.shape[0] - 1
|
|
875
1318
|
else:
|
|
876
|
-
l__dx=dx
|
|
1319
|
+
l__dx = dx
|
|
877
1320
|
|
|
878
|
-
if i__y.shape[0]<dell:
|
|
879
|
-
l__dell=0
|
|
1321
|
+
if i__y.shape[0] < dell:
|
|
1322
|
+
l__dell = 0
|
|
880
1323
|
else:
|
|
881
|
-
l__dell=dell
|
|
1324
|
+
l__dell = dell
|
|
882
1325
|
|
|
883
|
-
if l__dx<2:
|
|
884
|
-
res=np.zeros([i__y.shape[0]+add])
|
|
1326
|
+
if l__dx < 2:
|
|
1327
|
+
res = np.zeros([i__y.shape[0] + add])
|
|
885
1328
|
if inverse:
|
|
886
|
-
res[:-add]=i__y
|
|
1329
|
+
res[:-add] = i__y
|
|
887
1330
|
else:
|
|
888
|
-
res[add:]=i__y[0:]
|
|
1331
|
+
res[add:] = i__y[0:]
|
|
889
1332
|
return res
|
|
890
1333
|
|
|
891
1334
|
if weigth is None:
|
|
892
|
-
w=2**(np.arange(l__dx))
|
|
1335
|
+
w = 2 ** (np.arange(l__dx))
|
|
893
1336
|
else:
|
|
894
1337
|
if not inverse:
|
|
895
|
-
w=weigth[0:l__dx]
|
|
1338
|
+
w = weigth[0:l__dx]
|
|
896
1339
|
else:
|
|
897
|
-
w=weigth[-l__dx:]
|
|
1340
|
+
w = weigth[-l__dx:]
|
|
898
1341
|
|
|
899
|
-
x=np.arange(l__dx)+1
|
|
1342
|
+
x = np.arange(l__dx) + 1
|
|
900
1343
|
if not inverse:
|
|
901
|
-
y=np.log(i__y[1:l__dx+1])
|
|
1344
|
+
y = np.log(i__y[1 : l__dx + 1])
|
|
902
1345
|
else:
|
|
903
|
-
y=np.log(i__y[-(l__dx+1)
|
|
1346
|
+
y = np.log(i__y[-(l__dx + 1) : -1])
|
|
904
1347
|
|
|
905
|
-
r=np.polyfit(x,y,1,w=w)
|
|
1348
|
+
r = np.polyfit(x, y, 1, w=w)
|
|
906
1349
|
|
|
907
1350
|
if inverse:
|
|
908
|
-
res=np.exp(r[0]*(np.arange(i__y.shape[0]+add)-1)+r[1])
|
|
909
|
-
res[
|
|
1351
|
+
res = np.exp(r[0] * (np.arange(i__y.shape[0] + add) - 1) + r[1])
|
|
1352
|
+
res[: -(l__dell + add)] = i__y[:-l__dell]
|
|
910
1353
|
else:
|
|
911
|
-
res=np.exp(r[0]*(np.arange(i__y.shape[0]+add)-add)+r[1])
|
|
912
|
-
res[l__dell+add:]=i__y[l__dell:]
|
|
1354
|
+
res = np.exp(r[0] * (np.arange(i__y.shape[0] + add) - add) + r[1])
|
|
1355
|
+
res[l__dell + add :] = i__y[l__dell:]
|
|
913
1356
|
return res
|
|
914
1357
|
|
|
915
|
-
def findn(self,n):
|
|
916
|
-
d=np.sqrt(1+8*n)
|
|
917
|
-
return int((d-1)/2)
|
|
1358
|
+
def findn(self, n):
|
|
1359
|
+
d = np.sqrt(1 + 8 * n)
|
|
1360
|
+
return int((d - 1) / 2)
|
|
918
1361
|
|
|
919
|
-
def findidx(self,s2):
|
|
920
|
-
i1=np.zeros([s2.shape[1]],dtype=
|
|
921
|
-
i2=np.zeros([s2.shape[1]],dtype=
|
|
922
|
-
n=0
|
|
1362
|
+
def findidx(self, s2):
|
|
1363
|
+
i1 = np.zeros([s2.shape[1]], dtype="int")
|
|
1364
|
+
i2 = np.zeros([s2.shape[1]], dtype="int")
|
|
1365
|
+
n = 0
|
|
923
1366
|
for k in range(self.findn(s2.shape[1])):
|
|
924
|
-
i1[n:n+k+1]=np.arange(k+1)
|
|
925
|
-
i2[n:n+k+1]=k
|
|
926
|
-
n=n+k+1
|
|
927
|
-
return i1,i2
|
|
928
|
-
|
|
929
|
-
def extrapol_s2(self,add,lnorm=1):
|
|
930
|
-
if lnorm==1:
|
|
931
|
-
s2=self.S2.numpy()
|
|
932
|
-
if lnorm==2:
|
|
933
|
-
s2=self.S2L.numpy()
|
|
934
|
-
i1,i2=self.findidx(s2)
|
|
935
|
-
|
|
936
|
-
so2=np.zeros(
|
|
937
|
-
|
|
938
|
-
|
|
1367
|
+
i1[n : n + k + 1] = np.arange(k + 1)
|
|
1368
|
+
i2[n : n + k + 1] = k
|
|
1369
|
+
n = n + k + 1
|
|
1370
|
+
return i1, i2
|
|
1371
|
+
|
|
1372
|
+
def extrapol_s2(self, add, lnorm=1):
|
|
1373
|
+
if lnorm == 1:
|
|
1374
|
+
s2 = self.S2.numpy()
|
|
1375
|
+
if lnorm == 2:
|
|
1376
|
+
s2 = self.S2L.numpy()
|
|
1377
|
+
i1, i2 = self.findidx(s2)
|
|
1378
|
+
|
|
1379
|
+
so2 = np.zeros(
|
|
1380
|
+
[
|
|
1381
|
+
s2.shape[0],
|
|
1382
|
+
(self.findn(s2.shape[1]) + add)
|
|
1383
|
+
* (self.findn(s2.shape[1]) + add + 1)
|
|
1384
|
+
// 2,
|
|
1385
|
+
s2.shape[2],
|
|
1386
|
+
s2.shape[3],
|
|
1387
|
+
]
|
|
1388
|
+
)
|
|
1389
|
+
oi1, oi2 = self.findidx(so2)
|
|
1390
|
+
for l_batch in range(s2.shape[0]):
|
|
939
1391
|
for k in range(self.findn(s2.shape[1])):
|
|
940
1392
|
for i in range(s2.shape[2]):
|
|
941
1393
|
for j in range(s2.shape[3]):
|
|
942
|
-
tmp=self.model(
|
|
943
|
-
|
|
944
|
-
|
|
945
|
-
|
|
946
|
-
|
|
947
|
-
|
|
948
|
-
|
|
949
|
-
|
|
950
|
-
|
|
1394
|
+
tmp = self.model(
|
|
1395
|
+
s2[l_batch, i2 == k, i, j],
|
|
1396
|
+
dx=4,
|
|
1397
|
+
dell=1,
|
|
1398
|
+
add=add,
|
|
1399
|
+
weigth=np.array([1, 2, 2, 2]),
|
|
1400
|
+
)
|
|
1401
|
+
tmp[np.isnan(tmp)] = 0.0
|
|
1402
|
+
so2[l_batch, oi2 == k + add, i, j] = tmp
|
|
1403
|
+
|
|
1404
|
+
for l_batch in range(s2.shape[0]):
|
|
1405
|
+
for k in range(add + 1, -1, -1):
|
|
1406
|
+
lidx = np.where(oi2 - oi1 == k)[0]
|
|
1407
|
+
lidx2 = np.where(oi2 - oi1 == k + 1)[0]
|
|
951
1408
|
for i in range(s2.shape[2]):
|
|
952
1409
|
for j in range(s2.shape[3]):
|
|
953
|
-
so2[
|
|
1410
|
+
so2[l_batch, lidx[0 : add + 2 - k], i, j] = so2[
|
|
1411
|
+
l_batch, lidx2[0 : add + 2 - k], i, j
|
|
1412
|
+
]
|
|
954
1413
|
|
|
955
|
-
return
|
|
1414
|
+
return so2
|
|
956
1415
|
|
|
957
|
-
def extrapol_s1(self,i_s1,add):
|
|
958
|
-
s1=i_s1.numpy()
|
|
959
|
-
so1=np.zeros([s1.shape[0],s1.shape[1]+add,s1.shape[2]])
|
|
1416
|
+
def extrapol_s1(self, i_s1, add):
|
|
1417
|
+
s1 = i_s1.numpy()
|
|
1418
|
+
so1 = np.zeros([s1.shape[0], s1.shape[1] + add, s1.shape[2]])
|
|
960
1419
|
for k in range(s1.shape[0]):
|
|
961
1420
|
for i in range(s1.shape[2]):
|
|
962
|
-
so1[k
|
|
963
|
-
so1[k,np.isnan(so1[k
|
|
1421
|
+
so1[k, :, i] = self.model(s1[k, :, i], dx=4, dell=1, add=add)
|
|
1422
|
+
so1[k, np.isnan(so1[k, :, i]), i] = 0.0
|
|
964
1423
|
return so1
|
|
965
1424
|
|
|
966
|
-
def extrapol(self,add):
|
|
967
|
-
return scat(
|
|
968
|
-
|
|
969
|
-
|
|
970
|
-
|
|
971
|
-
|
|
972
|
-
|
|
973
|
-
|
|
974
|
-
|
|
975
|
-
|
|
976
|
-
|
|
1425
|
+
def extrapol(self, add):
|
|
1426
|
+
return scat(
|
|
1427
|
+
self.extrapol_s1(self.P00, add),
|
|
1428
|
+
self.S0,
|
|
1429
|
+
self.extrapol_s1(self.S1, add),
|
|
1430
|
+
self.extrapol_s2(add, lnorm=1),
|
|
1431
|
+
self.extrapol_s2(add, lnorm=2),
|
|
1432
|
+
self.j1,
|
|
1433
|
+
self.j2,
|
|
1434
|
+
backend=self.backend,
|
|
1435
|
+
)
|
|
1436
|
+
|
|
1437
|
+
|
|
977
1438
|
class funct(FOC.FoCUS):
|
|
978
|
-
|
|
979
|
-
def fill(self,im,nullval=hp.UNSEEN):
|
|
980
|
-
return self.fill_healpy(im,nullval=nullval)
|
|
981
|
-
|
|
982
|
-
def moments(self,list_scat):
|
|
983
|
-
S0=None
|
|
1439
|
+
|
|
1440
|
+
def fill(self, im, nullval=hp.UNSEEN):
|
|
1441
|
+
return self.fill_healpy(im, nullval=nullval)
|
|
1442
|
+
|
|
1443
|
+
def moments(self, list_scat):
|
|
1444
|
+
S0 = None
|
|
984
1445
|
for k in list_scat:
|
|
985
|
-
tmp=list_scat[k]
|
|
986
|
-
nS0=np.expand_dims(tmp.S0.numpy(),0)
|
|
987
|
-
nP00=np.expand_dims(tmp.P00.numpy(),0)
|
|
988
|
-
nS1=np.expand_dims(tmp.S1.numpy(),0)
|
|
989
|
-
nS2=np.expand_dims(tmp.S2.numpy(),0)
|
|
990
|
-
nS2L=np.expand_dims(tmp.S2L.numpy(),0)
|
|
991
|
-
|
|
1446
|
+
tmp = list_scat[k]
|
|
1447
|
+
nS0 = np.expand_dims(tmp.S0.numpy(), 0)
|
|
1448
|
+
nP00 = np.expand_dims(tmp.P00.numpy(), 0)
|
|
1449
|
+
nS1 = np.expand_dims(tmp.S1.numpy(), 0)
|
|
1450
|
+
nS2 = np.expand_dims(tmp.S2.numpy(), 0)
|
|
1451
|
+
nS2L = np.expand_dims(tmp.S2L.numpy(), 0)
|
|
1452
|
+
|
|
992
1453
|
if S0 is None:
|
|
993
|
-
S0=nS0
|
|
994
|
-
P00=nP00
|
|
995
|
-
S1=nS1
|
|
996
|
-
S2=nS2
|
|
997
|
-
S2L=nS2L
|
|
1454
|
+
S0 = nS0
|
|
1455
|
+
P00 = nP00
|
|
1456
|
+
S1 = nS1
|
|
1457
|
+
S2 = nS2
|
|
1458
|
+
S2L = nS2L
|
|
998
1459
|
else:
|
|
999
|
-
S0=np.concatenate([S0,nS0],0)
|
|
1000
|
-
P00=np.concatenate([P00,nP00],0)
|
|
1001
|
-
S1=np.concatenate([S1,nS1],0)
|
|
1002
|
-
S2=np.concatenate([S2,nS2],0)
|
|
1003
|
-
S2L=np.concatenate([S2L,nS2L],0)
|
|
1004
|
-
|
|
1005
|
-
sS0=np.std(S0,0)
|
|
1006
|
-
sP00=np.std(P00,0)
|
|
1007
|
-
sS1=np.std(S1,0)
|
|
1008
|
-
sS2=np.std(S2,0)
|
|
1009
|
-
sS2L=np.std(S2L,0)
|
|
1010
|
-
|
|
1011
|
-
mS0=np.mean(S0,0)
|
|
1012
|
-
mP00=np.mean(P00,0)
|
|
1013
|
-
mS1=np.mean(S1,0)
|
|
1014
|
-
mS2=np.mean(S2,0)
|
|
1015
|
-
mS2L=np.mean(S2L,0)
|
|
1016
|
-
|
|
1017
|
-
return scat(
|
|
1018
|
-
|
|
1019
|
-
|
|
1020
|
-
|
|
1460
|
+
S0 = np.concatenate([S0, nS0], 0)
|
|
1461
|
+
P00 = np.concatenate([P00, nP00], 0)
|
|
1462
|
+
S1 = np.concatenate([S1, nS1], 0)
|
|
1463
|
+
S2 = np.concatenate([S2, nS2], 0)
|
|
1464
|
+
S2L = np.concatenate([S2L, nS2L], 0)
|
|
1465
|
+
|
|
1466
|
+
sS0 = np.std(S0, 0)
|
|
1467
|
+
sP00 = np.std(P00, 0)
|
|
1468
|
+
sS1 = np.std(S1, 0)
|
|
1469
|
+
sS2 = np.std(S2, 0)
|
|
1470
|
+
sS2L = np.std(S2L, 0)
|
|
1471
|
+
|
|
1472
|
+
mS0 = np.mean(S0, 0)
|
|
1473
|
+
mP00 = np.mean(P00, 0)
|
|
1474
|
+
mS1 = np.mean(S1, 0)
|
|
1475
|
+
mS2 = np.mean(S2, 0)
|
|
1476
|
+
mS2L = np.mean(S2L, 0)
|
|
1477
|
+
|
|
1478
|
+
return scat(
|
|
1479
|
+
mP00, mS0, mS1, mS2, mS2L, tmp.j1, tmp.j2, backend=self.backend
|
|
1480
|
+
), scat(sP00, sS0, sS1, sS2, sS2L, tmp.j1, tmp.j2, backend=self.backend)
|
|
1481
|
+
|
|
1482
|
+
def eval(
|
|
1483
|
+
self,
|
|
1484
|
+
image1,
|
|
1485
|
+
image2=None,
|
|
1486
|
+
mask=None,
|
|
1487
|
+
Auto=True,
|
|
1488
|
+
s0_off=1e-6,
|
|
1489
|
+
calc_var=False,
|
|
1490
|
+
norm=None,
|
|
1491
|
+
):
|
|
1021
1492
|
# Check input consistency
|
|
1022
1493
|
if image2 is not None:
|
|
1023
|
-
if list(image1.shape)!=list(image2.shape):
|
|
1024
|
-
print(
|
|
1025
|
-
|
|
1026
|
-
|
|
1494
|
+
if list(image1.shape) != list(image2.shape):
|
|
1495
|
+
print(
|
|
1496
|
+
"The two input image should have the same size to eval Scattering"
|
|
1497
|
+
)
|
|
1498
|
+
|
|
1499
|
+
return None
|
|
1027
1500
|
if mask is not None:
|
|
1028
|
-
if list(image1.shape)!=list(mask.shape)[1:]:
|
|
1029
|
-
print(
|
|
1030
|
-
|
|
1031
|
-
|
|
1032
|
-
|
|
1033
|
-
|
|
1034
|
-
|
|
1035
|
-
|
|
1036
|
-
|
|
1501
|
+
if list(image1.shape) != list(mask.shape)[1:]:
|
|
1502
|
+
print(
|
|
1503
|
+
"The mask should have the same size than the input image to eval Scattering"
|
|
1504
|
+
)
|
|
1505
|
+
print("Image shape ", image1.shape, "Mask shape ", mask.shape)
|
|
1506
|
+
return None
|
|
1507
|
+
if self.use_2D and len(image1.shape) < 2:
|
|
1508
|
+
print(
|
|
1509
|
+
"To work with 2D scattering transform, two dimension is needed, input map has only on dimension"
|
|
1510
|
+
)
|
|
1511
|
+
return None
|
|
1512
|
+
|
|
1037
1513
|
### AUTO OR CROSS
|
|
1038
1514
|
cross = False
|
|
1039
1515
|
if image2 is not None:
|
|
1040
1516
|
cross = True
|
|
1041
|
-
|
|
1042
|
-
else:
|
|
1043
|
-
all_cross=False
|
|
1044
|
-
|
|
1517
|
+
|
|
1045
1518
|
# Check if image1 is [Npix] or [Nbatch,Npix]
|
|
1046
|
-
axis=1
|
|
1047
|
-
|
|
1519
|
+
axis = 1
|
|
1520
|
+
|
|
1048
1521
|
# determine jmax and nside corresponding to the input map
|
|
1049
1522
|
im_shape = image1.shape
|
|
1050
1523
|
if self.use_2D:
|
|
1051
|
-
if len(image1.shape)==2:
|
|
1052
|
-
nside=np.min([im_shape[0],im_shape[1]])
|
|
1053
|
-
npix = im_shape[0]*im_shape[1]
|
|
1054
|
-
x1=im_shape[0]
|
|
1055
|
-
x2=im_shape[1]
|
|
1524
|
+
if len(image1.shape) == 2:
|
|
1525
|
+
nside = np.min([im_shape[0], im_shape[1]])
|
|
1526
|
+
npix = im_shape[0] * im_shape[1] # Number of pixels
|
|
1056
1527
|
else:
|
|
1057
|
-
nside=np.min([im_shape[1],im_shape[2]])
|
|
1058
|
-
npix = im_shape[1]*im_shape[2]
|
|
1059
|
-
|
|
1060
|
-
x2=im_shape[2]
|
|
1061
|
-
jmax = int(np.log(nside-self.KERNELSZ) / np.log(2)) # Number of j scales
|
|
1528
|
+
nside = np.min([im_shape[1], im_shape[2]])
|
|
1529
|
+
npix = im_shape[1] * im_shape[2] # Number of pixels
|
|
1530
|
+
jmax = int(np.log(nside - self.KERNELSZ) / np.log(2)) # Number of j scales
|
|
1062
1531
|
else:
|
|
1063
|
-
if len(image1.shape)==2:
|
|
1532
|
+
if len(image1.shape) == 2:
|
|
1064
1533
|
npix = int(im_shape[1]) # Number of pixels
|
|
1065
1534
|
else:
|
|
1066
1535
|
npix = int(im_shape[0]) # Number of pixels
|
|
1067
1536
|
|
|
1068
|
-
nside=int(np.sqrt(npix//12))
|
|
1069
|
-
|
|
1070
|
-
jmax=int(np.log(nside)/np.log(2))
|
|
1537
|
+
nside = int(np.sqrt(npix // 12))
|
|
1538
|
+
|
|
1539
|
+
jmax = int(np.log(nside) / np.log(2)) # -self.OSTEP
|
|
1071
1540
|
|
|
1072
1541
|
### LOCAL VARIABLES (IMAGES and MASK)
|
|
1073
1542
|
# Check if image1 is [Npix] or [Nbatch,Npix]
|
|
1074
|
-
if len(image1.shape)==1 or (len(image1.shape)==2 and self.use_2D):
|
|
1543
|
+
if len(image1.shape) == 1 or (len(image1.shape) == 2 and self.use_2D):
|
|
1075
1544
|
# image1 is [Nbatch, Npix]
|
|
1076
|
-
I1 = self.backend.bk_cast(
|
|
1545
|
+
I1 = self.backend.bk_cast(
|
|
1546
|
+
self.backend.bk_expand_dims(image1, 0)
|
|
1547
|
+
) # Local image1 [Nbatch, Npix]
|
|
1077
1548
|
if cross:
|
|
1078
|
-
I2 = self.backend.bk_cast(
|
|
1549
|
+
I2 = self.backend.bk_cast(
|
|
1550
|
+
self.backend.bk_expand_dims(image2, 0)
|
|
1551
|
+
) # Local image2 [Nbatch, Npix]
|
|
1079
1552
|
else:
|
|
1080
|
-
I1=self.backend.bk_cast(image1)
|
|
1553
|
+
I1 = self.backend.bk_cast(image1)
|
|
1081
1554
|
if cross:
|
|
1082
|
-
I2=self.backend.bk_cast(image2)
|
|
1083
|
-
|
|
1555
|
+
I2 = self.backend.bk_cast(image2)
|
|
1556
|
+
|
|
1084
1557
|
# self.mask is [Nmask, Npix]
|
|
1085
1558
|
if mask is None:
|
|
1086
1559
|
if self.use_2D:
|
|
1087
|
-
vmask = self.backend.bk_ones(
|
|
1560
|
+
vmask = self.backend.bk_ones(
|
|
1561
|
+
[1, I1.shape[axis], I1.shape[axis + 1]], dtype=self.all_type
|
|
1562
|
+
)
|
|
1088
1563
|
else:
|
|
1089
1564
|
vmask = self.backend.bk_ones([1, I1.shape[axis]], dtype=self.all_type)
|
|
1090
1565
|
else:
|
|
1091
1566
|
vmask = self.backend.bk_cast(mask) # [Nmask, Npix]
|
|
1092
1567
|
|
|
1093
|
-
if self.KERNELSZ>3:
|
|
1094
|
-
if self.KERNELSZ==5:
|
|
1568
|
+
if self.KERNELSZ > 3:
|
|
1569
|
+
if self.KERNELSZ == 5:
|
|
1095
1570
|
# if the kernel size is bigger than 3 increase the binning before smoothing
|
|
1096
1571
|
if self.use_2D:
|
|
1097
|
-
l_image1=self.up_grade(
|
|
1098
|
-
|
|
1572
|
+
l_image1 = self.up_grade(
|
|
1573
|
+
I1, I1.shape[axis] * 2, axis=axis, nouty=I1.shape[axis + 1] * 2
|
|
1574
|
+
)
|
|
1575
|
+
vmask = self.up_grade(
|
|
1576
|
+
vmask, I1.shape[axis] * 2, axis=1, nouty=I1.shape[axis + 1] * 2
|
|
1577
|
+
)
|
|
1099
1578
|
else:
|
|
1100
|
-
l_image1=self.up_grade(I1,nside*2,axis=axis)
|
|
1101
|
-
vmask=self.up_grade(vmask,nside*2,axis=1)
|
|
1102
|
-
|
|
1579
|
+
l_image1 = self.up_grade(I1, nside * 2, axis=axis)
|
|
1580
|
+
vmask = self.up_grade(vmask, nside * 2, axis=1)
|
|
1581
|
+
|
|
1103
1582
|
if cross:
|
|
1104
1583
|
if self.use_2D:
|
|
1105
|
-
l_image2=self.up_grade(
|
|
1584
|
+
l_image2 = self.up_grade(
|
|
1585
|
+
I2,
|
|
1586
|
+
I2.shape[axis] * 2,
|
|
1587
|
+
axis=axis,
|
|
1588
|
+
nouty=I2.shape[axis + 1] * 2,
|
|
1589
|
+
)
|
|
1106
1590
|
else:
|
|
1107
|
-
l_image2=self.up_grade(I2,nside*2,axis=axis)
|
|
1591
|
+
l_image2 = self.up_grade(I2, nside * 2, axis=axis)
|
|
1108
1592
|
else:
|
|
1109
1593
|
# if the kernel size is bigger than 3 increase the binning before smoothing
|
|
1110
1594
|
if self.use_2D:
|
|
1111
|
-
|
|
1112
|
-
|
|
1113
|
-
|
|
1595
|
+
l_image1 = self.up_grade(
|
|
1596
|
+
l_image1,
|
|
1597
|
+
I1.shape[axis] * 4,
|
|
1598
|
+
axis=axis,
|
|
1599
|
+
nouty=I1.shape[axis + 1] * 4,
|
|
1600
|
+
)
|
|
1601
|
+
vmask = self.up_grade(
|
|
1602
|
+
vmask, I1.shape[axis] * 4, axis=1, nouty=I1.shape[axis + 1] * 4
|
|
1603
|
+
)
|
|
1114
1604
|
else:
|
|
1115
|
-
l_image1=self.up_grade(l_image1,nside*4,axis=axis)
|
|
1116
|
-
vmask=self.up_grade(vmask,nside*4,axis=1)
|
|
1117
|
-
|
|
1605
|
+
l_image1 = self.up_grade(l_image1, nside * 4, axis=axis)
|
|
1606
|
+
vmask = self.up_grade(vmask, nside * 4, axis=1)
|
|
1607
|
+
|
|
1118
1608
|
if cross:
|
|
1119
1609
|
if self.use_2D:
|
|
1120
|
-
l_image2=self.up_grade(
|
|
1610
|
+
l_image2 = self.up_grade(
|
|
1611
|
+
l_image2,
|
|
1612
|
+
I2.shape[axis] * 4,
|
|
1613
|
+
axis=axis,
|
|
1614
|
+
nouty=I2.shape[axis + 1] * 4,
|
|
1615
|
+
)
|
|
1121
1616
|
else:
|
|
1122
|
-
l_image2=self.up_grade(l_image2,nside*4,axis=axis)
|
|
1617
|
+
l_image2 = self.up_grade(l_image2, nside * 4, axis=axis)
|
|
1123
1618
|
else:
|
|
1124
|
-
l_image1=I1
|
|
1619
|
+
l_image1 = I1
|
|
1125
1620
|
if cross:
|
|
1126
|
-
l_image2=I2
|
|
1621
|
+
l_image2 = I2
|
|
1127
1622
|
|
|
1128
1623
|
if calc_var:
|
|
1129
|
-
s0,vs0 = self.masked_mean(l_image1,vmask,axis=axis,calc_var=True)
|
|
1130
|
-
s0=s0+s0_off
|
|
1624
|
+
s0, vs0 = self.masked_mean(l_image1, vmask, axis=axis, calc_var=True)
|
|
1625
|
+
s0 = s0 + s0_off
|
|
1131
1626
|
else:
|
|
1132
|
-
s0 = self.masked_mean(l_image1,vmask,axis=axis)+s0_off
|
|
1133
|
-
|
|
1134
|
-
if cross and Auto
|
|
1627
|
+
s0 = self.masked_mean(l_image1, vmask, axis=axis) + s0_off
|
|
1628
|
+
|
|
1629
|
+
if cross and not Auto:
|
|
1135
1630
|
if calc_var:
|
|
1136
|
-
s02,vs02=self.masked_mean(l_image2,vmask,axis=axis,calc_var=True)
|
|
1631
|
+
s02, vs02 = self.masked_mean(l_image2, vmask, axis=axis, calc_var=True)
|
|
1137
1632
|
else:
|
|
1138
|
-
s02=self.masked_mean(l_image2,vmask,axis=axis)
|
|
1139
|
-
|
|
1140
|
-
if len(image1.shape)==1 or (len(image1.shape)==2 and self.use_2D):
|
|
1633
|
+
s02 = self.masked_mean(l_image2, vmask, axis=axis)
|
|
1634
|
+
|
|
1635
|
+
if len(image1.shape) == 1 or (len(image1.shape) == 2 and self.use_2D):
|
|
1141
1636
|
if self.backend.bk_is_complex(s0):
|
|
1142
|
-
s0 = self.backend.bk_complex(s0,s02+s0_off)
|
|
1637
|
+
s0 = self.backend.bk_complex(s0, s02 + s0_off)
|
|
1143
1638
|
if calc_var:
|
|
1144
|
-
vs0 = self.backend.bk_complex(vs0,vs02)
|
|
1639
|
+
vs0 = self.backend.bk_complex(vs0, vs02)
|
|
1145
1640
|
else:
|
|
1146
|
-
s0 = self.backend.bk_concat([s0,s02],axis=0)
|
|
1641
|
+
s0 = self.backend.bk_concat([s0, s02], axis=0)
|
|
1147
1642
|
if calc_var:
|
|
1148
|
-
vs0 = self.backend.bk_concat([vs0,vs02],axis=0)
|
|
1643
|
+
vs0 = self.backend.bk_concat([vs0, vs02], axis=0)
|
|
1149
1644
|
else:
|
|
1150
1645
|
if self.backend.bk_is_complex(s0):
|
|
1151
|
-
s0 = self.backend.bk_complex(s0,s02+s0_off)
|
|
1646
|
+
s0 = self.backend.bk_complex(s0, s02 + s0_off)
|
|
1152
1647
|
if calc_var:
|
|
1153
|
-
vs0 = self.backend.bk_complex(vs0,vs02)
|
|
1648
|
+
vs0 = self.backend.bk_complex(vs0, vs02)
|
|
1154
1649
|
else:
|
|
1155
|
-
s0 = self.backend.bk_concat([s0,s02],axis=0)
|
|
1650
|
+
s0 = self.backend.bk_concat([s0, s02], axis=0)
|
|
1156
1651
|
if calc_var:
|
|
1157
|
-
vs0 = self.backend.bk_concat([vs0,vs02],axis=0)
|
|
1158
|
-
|
|
1159
|
-
s1=None
|
|
1160
|
-
s2=None
|
|
1161
|
-
s2l=None
|
|
1162
|
-
p00=None
|
|
1163
|
-
s2j1=None
|
|
1164
|
-
s2j2=None
|
|
1165
|
-
|
|
1166
|
-
l2_image=None
|
|
1167
|
-
l2_image_imag=None
|
|
1168
|
-
|
|
1652
|
+
vs0 = self.backend.bk_concat([vs0, vs02], axis=0)
|
|
1653
|
+
|
|
1654
|
+
s1 = None
|
|
1655
|
+
s2 = None
|
|
1656
|
+
s2l = None
|
|
1657
|
+
p00 = None
|
|
1658
|
+
s2j1 = None
|
|
1659
|
+
s2j2 = None
|
|
1660
|
+
l2_image = None
|
|
1169
1661
|
for j1 in range(jmax):
|
|
1170
|
-
if j1<jmax-self.OSTEP:
|
|
1662
|
+
if j1 < jmax - self.OSTEP: # stop to add scales
|
|
1171
1663
|
# Convol image along the axis defined by 'axis' using the wavelet defined at
|
|
1172
1664
|
# the foscat initialisation
|
|
1173
|
-
#c_image_real is [....,Npix_j1,....,Norient]
|
|
1174
|
-
c_image1=self.convol(l_image1,axis=axis)
|
|
1665
|
+
# c_image_real is [....,Npix_j1,....,Norient]
|
|
1666
|
+
c_image1 = self.convol(l_image1, axis=axis)
|
|
1175
1667
|
if cross:
|
|
1176
|
-
c_image2=self.convol(l_image2,axis=axis)
|
|
1668
|
+
c_image2 = self.convol(l_image2, axis=axis)
|
|
1177
1669
|
else:
|
|
1178
|
-
c_image2=c_image1
|
|
1670
|
+
c_image2 = c_image1
|
|
1179
1671
|
|
|
1180
1672
|
# Compute (a+ib)*(a+ib)* the last c_image column is the real and imaginary part
|
|
1181
|
-
conj=c_image1*self.backend.bk_conjugate(c_image2)
|
|
1182
|
-
|
|
1673
|
+
conj = c_image1 * self.backend.bk_conjugate(c_image2)
|
|
1674
|
+
|
|
1183
1675
|
if Auto:
|
|
1184
|
-
conj=self.backend.bk_real(conj)
|
|
1676
|
+
conj = self.backend.bk_real(conj)
|
|
1185
1677
|
|
|
1186
1678
|
# Compute l_p00 [....,....,Nmask,j1,Norient]
|
|
1187
1679
|
if calc_var:
|
|
1188
|
-
l_p00,l_vp00 = self.masked_mean(
|
|
1189
|
-
|
|
1190
|
-
|
|
1680
|
+
l_p00, l_vp00 = self.masked_mean(
|
|
1681
|
+
conj, vmask, axis=axis, rank=j1, calc_var=True
|
|
1682
|
+
)
|
|
1683
|
+
l_p00 = self.backend.bk_expand_dims(l_p00, -2)
|
|
1684
|
+
l_vp00 = self.backend.bk_expand_dims(l_vp00, -2)
|
|
1191
1685
|
else:
|
|
1192
|
-
l_p00 = self.masked_mean(conj,vmask,axis=axis,rank=j1)
|
|
1193
|
-
l_p00 = self.backend.bk_expand_dims(l_p00
|
|
1686
|
+
l_p00 = self.masked_mean(conj, vmask, axis=axis, rank=j1)
|
|
1687
|
+
l_p00 = self.backend.bk_expand_dims(l_p00, -2)
|
|
1194
1688
|
|
|
1195
|
-
conj=self.backend.bk_L1(conj)
|
|
1689
|
+
conj = self.backend.bk_L1(conj)
|
|
1196
1690
|
|
|
1197
|
-
# Compute l_s1 [....,....,Nmask,1,Norient]
|
|
1691
|
+
# Compute l_s1 [....,....,Nmask,1,Norient]
|
|
1198
1692
|
if calc_var:
|
|
1199
|
-
l_s1,l_vs1 = self.masked_mean(
|
|
1200
|
-
|
|
1201
|
-
|
|
1693
|
+
l_s1, l_vs1 = self.masked_mean(
|
|
1694
|
+
conj, vmask, axis=axis, rank=j1, calc_var=True
|
|
1695
|
+
)
|
|
1696
|
+
l_s1 = self.backend.bk_expand_dims(l_s1, -2)
|
|
1697
|
+
l_vs1 = self.backend.bk_expand_dims(l_vs1, -2)
|
|
1202
1698
|
else:
|
|
1203
|
-
l_s1 = self.backend.bk_expand_dims(
|
|
1699
|
+
l_s1 = self.backend.bk_expand_dims(
|
|
1700
|
+
self.masked_mean(conj, vmask, axis=axis, rank=j1), -2
|
|
1701
|
+
)
|
|
1204
1702
|
|
|
1205
|
-
# Concat S1,P00 [....,....,Nmask,j1,Norient]
|
|
1703
|
+
# Concat S1,P00 [....,....,Nmask,j1,Norient]
|
|
1206
1704
|
if s1 is None:
|
|
1207
|
-
s1=l_s1
|
|
1208
|
-
p00=l_p00
|
|
1705
|
+
s1 = l_s1
|
|
1706
|
+
p00 = l_p00
|
|
1209
1707
|
if calc_var:
|
|
1210
|
-
vs1=l_vs1
|
|
1211
|
-
vp00=l_vp00
|
|
1708
|
+
vs1 = l_vs1
|
|
1709
|
+
vp00 = l_vp00
|
|
1212
1710
|
else:
|
|
1213
|
-
s1=self.backend.bk_concat([s1,l_s1],axis=-2)
|
|
1214
|
-
p00=self.backend.bk_concat([p00,l_p00],axis=-2)
|
|
1711
|
+
s1 = self.backend.bk_concat([s1, l_s1], axis=-2)
|
|
1712
|
+
p00 = self.backend.bk_concat([p00, l_p00], axis=-2)
|
|
1215
1713
|
if calc_var:
|
|
1216
|
-
vs1=self.backend.bk_concat([vs1,l_vs1],axis=-2)
|
|
1217
|
-
vp00=self.backend.bk_concat([vp00,l_vp00],axis=-2)
|
|
1714
|
+
vs1 = self.backend.bk_concat([vs1, l_vs1], axis=-2)
|
|
1715
|
+
vp00 = self.backend.bk_concat([vp00, l_vp00], axis=-2)
|
|
1218
1716
|
|
|
1219
1717
|
# Concat l2_image [....,j1,Npix_j1,,....,Norient]
|
|
1220
1718
|
if l2_image is None:
|
|
1221
1719
|
if self.use_2D:
|
|
1222
|
-
l2_image=self.backend.bk_expand_dims(conj,axis=-4)
|
|
1720
|
+
l2_image = self.backend.bk_expand_dims(conj, axis=-4)
|
|
1223
1721
|
else:
|
|
1224
|
-
l2_image=self.backend.bk_expand_dims(conj,axis=-3)
|
|
1722
|
+
l2_image = self.backend.bk_expand_dims(conj, axis=-3)
|
|
1225
1723
|
else:
|
|
1226
1724
|
if self.use_2D:
|
|
1227
|
-
l2_image=self.backend.bk_concat(
|
|
1725
|
+
l2_image = self.backend.bk_concat(
|
|
1726
|
+
[self.backend.bk_expand_dims(conj, axis=-4), l2_image],
|
|
1727
|
+
axis=-4,
|
|
1728
|
+
)
|
|
1228
1729
|
else:
|
|
1229
|
-
l2_image=self.backend.bk_concat(
|
|
1730
|
+
l2_image = self.backend.bk_concat(
|
|
1731
|
+
[self.backend.bk_expand_dims(conj, axis=-3), l2_image],
|
|
1732
|
+
axis=-3,
|
|
1733
|
+
)
|
|
1230
1734
|
|
|
1231
1735
|
# Convol l2_image [....,Npix_j1,j1,....,Norient,Norient]
|
|
1232
|
-
c2_image=self.convol(self.backend.bk_relu(l2_image),axis=axis+1)
|
|
1736
|
+
c2_image = self.convol(self.backend.bk_relu(l2_image), axis=axis + 1)
|
|
1233
1737
|
|
|
1234
|
-
conj2p=c2_image*self.backend.bk_conjugate(c2_image)
|
|
1235
|
-
conj2pl1=self.backend.bk_L1(conj2p)
|
|
1738
|
+
conj2p = c2_image * self.backend.bk_conjugate(c2_image)
|
|
1739
|
+
conj2pl1 = self.backend.bk_L1(conj2p)
|
|
1236
1740
|
|
|
1237
1741
|
if Auto:
|
|
1238
|
-
conj2p=self.backend.bk_real(conj2p)
|
|
1239
|
-
conj2pl1=self.backend.bk_real(conj2pl1)
|
|
1742
|
+
conj2p = self.backend.bk_real(conj2p)
|
|
1743
|
+
conj2pl1 = self.backend.bk_real(conj2pl1)
|
|
1240
1744
|
|
|
1241
|
-
c2_image=self.convol(self.backend.bk_relu(-l2_image),axis=axis+1)
|
|
1745
|
+
c2_image = self.convol(self.backend.bk_relu(-l2_image), axis=axis + 1)
|
|
1242
1746
|
|
|
1243
|
-
conj2m=c2_image*self.backend.bk_conjugate(c2_image)
|
|
1244
|
-
conj2ml1=self.backend.bk_L1(conj2m)
|
|
1747
|
+
conj2m = c2_image * self.backend.bk_conjugate(c2_image)
|
|
1748
|
+
conj2ml1 = self.backend.bk_L1(conj2m)
|
|
1245
1749
|
|
|
1246
1750
|
if Auto:
|
|
1247
|
-
conj2m=self.backend.bk_real(conj2m)
|
|
1248
|
-
conj2ml1=self.backend.bk_real(conj2ml1)
|
|
1249
|
-
|
|
1751
|
+
conj2m = self.backend.bk_real(conj2m)
|
|
1752
|
+
conj2ml1 = self.backend.bk_real(conj2ml1)
|
|
1753
|
+
|
|
1250
1754
|
# Convol l_s2 [....,....,Nmask,j1,Norient,Norient]
|
|
1251
1755
|
if calc_var:
|
|
1252
|
-
l_s2,l_vs2 = self.masked_mean(
|
|
1253
|
-
|
|
1756
|
+
l_s2, l_vs2 = self.masked_mean(
|
|
1757
|
+
conj2p - conj2m, vmask, axis=axis + 1, rank=j1, calc_var=True
|
|
1758
|
+
)
|
|
1759
|
+
l_s2l1, l_vs2l1 = self.masked_mean(
|
|
1760
|
+
conj2pl1 - conj2ml1, vmask, axis=axis + 1, rank=j1, calc_var=True
|
|
1761
|
+
)
|
|
1254
1762
|
else:
|
|
1255
|
-
l_s2 = self.masked_mean(conj2p-conj2m,vmask,axis=axis+1,rank=j1)
|
|
1256
|
-
l_s2l1 = self.masked_mean(
|
|
1763
|
+
l_s2 = self.masked_mean(conj2p - conj2m, vmask, axis=axis + 1, rank=j1)
|
|
1764
|
+
l_s2l1 = self.masked_mean(
|
|
1765
|
+
conj2pl1 - conj2ml1, vmask, axis=axis + 1, rank=j1
|
|
1766
|
+
)
|
|
1257
1767
|
|
|
1258
1768
|
# Concat l_s2 [....,....,Nmask,j1*(j1+1)/2,Norient,Norient]
|
|
1259
1769
|
if s2 is None:
|
|
1260
|
-
s2l=l_s2
|
|
1261
|
-
s2=l_s2l1
|
|
1770
|
+
s2l = l_s2
|
|
1771
|
+
s2 = l_s2l1
|
|
1262
1772
|
if calc_var:
|
|
1263
|
-
vs2l=l_vs2
|
|
1264
|
-
vs2=l_vs2l1
|
|
1265
|
-
|
|
1266
|
-
s2j1=np.arange(l_s2.shape[axis+1],dtype=
|
|
1267
|
-
s2j2=j1*np.ones(l_s2.shape[axis+1],dtype=
|
|
1773
|
+
vs2l = l_vs2
|
|
1774
|
+
vs2 = l_vs2l1
|
|
1775
|
+
|
|
1776
|
+
s2j1 = np.arange(l_s2.shape[axis + 1], dtype="int")
|
|
1777
|
+
s2j2 = j1 * np.ones(l_s2.shape[axis + 1], dtype="int")
|
|
1268
1778
|
else:
|
|
1269
|
-
s2=self.backend.bk_concat([s2,l_s2l1],axis=-3)
|
|
1270
|
-
s2l=self.backend.bk_concat([s2l,l_s2],axis=-3)
|
|
1779
|
+
s2 = self.backend.bk_concat([s2, l_s2l1], axis=-3)
|
|
1780
|
+
s2l = self.backend.bk_concat([s2l, l_s2], axis=-3)
|
|
1271
1781
|
if calc_var:
|
|
1272
|
-
vs2=self.backend.bk_concat([vs2,l_vs2l1],axis=-3)
|
|
1273
|
-
vs2l=self.backend.bk_concat([vs2l,l_vs2],axis=-3)
|
|
1274
|
-
|
|
1275
|
-
s2j1=np.concatenate(
|
|
1276
|
-
|
|
1277
|
-
|
|
1278
|
-
|
|
1279
|
-
|
|
1280
|
-
|
|
1281
|
-
|
|
1782
|
+
vs2 = self.backend.bk_concat([vs2, l_vs2l1], axis=-3)
|
|
1783
|
+
vs2l = self.backend.bk_concat([vs2l, l_vs2], axis=-3)
|
|
1784
|
+
|
|
1785
|
+
s2j1 = np.concatenate(
|
|
1786
|
+
[s2j1, np.arange(l_s2.shape[axis + 1], dtype="int")], 0
|
|
1787
|
+
)
|
|
1788
|
+
s2j2 = np.concatenate(
|
|
1789
|
+
[s2j2, j1 * np.ones(l_s2.shape[axis + 1], dtype="int")], 0
|
|
1790
|
+
)
|
|
1791
|
+
|
|
1792
|
+
if j1 != jmax - 1:
|
|
1793
|
+
# Rescale vmask [Nmask,Npix_j1//4]
|
|
1794
|
+
vmask = self.smooth(vmask, axis=1)
|
|
1795
|
+
vmask = self.ud_grade_2(vmask, axis=1)
|
|
1282
1796
|
if self.mask_thres is not None:
|
|
1283
|
-
vmask = self.backend.bk_threshold(vmask,self.mask_thres)
|
|
1797
|
+
vmask = self.backend.bk_threshold(vmask, self.mask_thres)
|
|
1284
1798
|
|
|
1285
|
-
# Rescale l2_image [....,Npix_j1//4,....,j1,Norient]
|
|
1286
|
-
l2_image = self.smooth(l2_image,axis=axis+1)
|
|
1287
|
-
l2_image = self.ud_grade_2(l2_image,axis=axis+1)
|
|
1799
|
+
# Rescale l2_image [....,Npix_j1//4,....,j1,Norient]
|
|
1800
|
+
l2_image = self.smooth(l2_image, axis=axis + 1)
|
|
1801
|
+
l2_image = self.ud_grade_2(l2_image, axis=axis + 1)
|
|
1288
1802
|
|
|
1289
|
-
# Rescale l_image [....,Npix_j1//4,....]
|
|
1290
|
-
l_image1 = self.smooth(l_image1,axis=axis)
|
|
1291
|
-
l_image1 = self.ud_grade_2(l_image1,axis=axis)
|
|
1803
|
+
# Rescale l_image [....,Npix_j1//4,....]
|
|
1804
|
+
l_image1 = self.smooth(l_image1, axis=axis)
|
|
1805
|
+
l_image1 = self.ud_grade_2(l_image1, axis=axis)
|
|
1292
1806
|
if cross:
|
|
1293
|
-
l_image2 = self.smooth(l_image2,axis=axis)
|
|
1294
|
-
l_image2 = self.ud_grade_2(l_image2,axis=axis)
|
|
1295
|
-
|
|
1296
|
-
|
|
1297
|
-
|
|
1298
|
-
|
|
1807
|
+
l_image2 = self.smooth(l_image2, axis=axis)
|
|
1808
|
+
l_image2 = self.ud_grade_2(l_image2, axis=axis)
|
|
1809
|
+
|
|
1810
|
+
if len(image1.shape) == 1 or (len(image1.shape) == 2 and self.use_2D):
|
|
1811
|
+
sc_ret = scat(
|
|
1812
|
+
p00[0],
|
|
1813
|
+
s0[0],
|
|
1814
|
+
s1[0],
|
|
1815
|
+
s2[0],
|
|
1816
|
+
s2l[0],
|
|
1817
|
+
s2j1,
|
|
1818
|
+
s2j2,
|
|
1819
|
+
cross=cross,
|
|
1820
|
+
backend=self.backend,
|
|
1821
|
+
)
|
|
1299
1822
|
else:
|
|
1300
|
-
sc_ret=scat(
|
|
1301
|
-
|
|
1823
|
+
sc_ret = scat(
|
|
1824
|
+
p00, s0, s1, s2, s2l, s2j1, s2j2, cross=cross, backend=self.backend
|
|
1825
|
+
)
|
|
1826
|
+
|
|
1302
1827
|
if calc_var:
|
|
1303
|
-
if len(image1.shape)==1 or (len(image1.shape)==2 and self.use_2D):
|
|
1304
|
-
vsc_ret=scat(
|
|
1828
|
+
if len(image1.shape) == 1 or (len(image1.shape) == 2 and self.use_2D):
|
|
1829
|
+
vsc_ret = scat(
|
|
1830
|
+
vp00[0],
|
|
1831
|
+
vs0[0],
|
|
1832
|
+
vs1[0],
|
|
1833
|
+
vs2[0],
|
|
1834
|
+
vs2l[0],
|
|
1835
|
+
s2j1,
|
|
1836
|
+
s2j2,
|
|
1837
|
+
cross=cross,
|
|
1838
|
+
backend=self.backend,
|
|
1839
|
+
)
|
|
1305
1840
|
else:
|
|
1306
|
-
vsc_ret=scat(
|
|
1307
|
-
|
|
1841
|
+
vsc_ret = scat(
|
|
1842
|
+
vp00,
|
|
1843
|
+
vs0,
|
|
1844
|
+
vs1,
|
|
1845
|
+
vs2,
|
|
1846
|
+
vs2l,
|
|
1847
|
+
s2j1,
|
|
1848
|
+
s2j2,
|
|
1849
|
+
cross=cross,
|
|
1850
|
+
backend=self.backend,
|
|
1851
|
+
)
|
|
1852
|
+
return sc_ret, vsc_ret
|
|
1308
1853
|
else:
|
|
1309
1854
|
return sc_ret
|
|
1310
1855
|
|
|
1311
|
-
def square(self,x):
|
|
1856
|
+
def square(self, x):
|
|
1312
1857
|
# the abs make the complex value usable for reduce_sum or mean
|
|
1313
|
-
return scat(
|
|
1314
|
-
|
|
1315
|
-
|
|
1316
|
-
|
|
1317
|
-
|
|
1318
|
-
|
|
1319
|
-
|
|
1858
|
+
return scat(
|
|
1859
|
+
self.backend.bk_square(self.backend.bk_abs(x.P00)),
|
|
1860
|
+
self.backend.bk_square(self.backend.bk_abs(x.S0)),
|
|
1861
|
+
self.backend.bk_square(self.backend.bk_abs(x.S1)),
|
|
1862
|
+
self.backend.bk_square(self.backend.bk_abs(x.S2)),
|
|
1863
|
+
self.backend.bk_square(self.backend.bk_abs(x.S2L)),
|
|
1864
|
+
x.j1,
|
|
1865
|
+
x.j2,
|
|
1866
|
+
backend=self.backend,
|
|
1867
|
+
)
|
|
1868
|
+
|
|
1869
|
+
def sqrt(self, x):
|
|
1320
1870
|
# the abs make the complex value usable for reduce_sum or mean
|
|
1321
|
-
return scat(
|
|
1322
|
-
|
|
1323
|
-
|
|
1324
|
-
|
|
1325
|
-
|
|
1871
|
+
return scat(
|
|
1872
|
+
self.backend.bk_sqrt(self.backend.bk_abs(x.P00)),
|
|
1873
|
+
self.backend.bk_sqrt(self.backend.bk_abs(x.S0)),
|
|
1874
|
+
self.backend.bk_sqrt(self.backend.bk_abs(x.S1)),
|
|
1875
|
+
self.backend.bk_sqrt(self.backend.bk_abs(x.S2)),
|
|
1876
|
+
self.backend.bk_sqrt(self.backend.bk_abs(x.S2L)),
|
|
1877
|
+
x.j1,
|
|
1878
|
+
x.j2,
|
|
1879
|
+
backend=self.backend,
|
|
1880
|
+
)
|
|
1881
|
+
|
|
1882
|
+
def reduce_distance(self, x, y, sigma=None):
|
|
1883
|
+
|
|
1884
|
+
if isinstance(x, scat):
|
|
1885
|
+
if sigma is None:
|
|
1886
|
+
result = self.diff_data(y.S0, x.S0, is_complex=False)
|
|
1887
|
+
result += self.diff_data(y.S1, x.S1)
|
|
1888
|
+
result += self.diff_data(y.P00, x.P00)
|
|
1889
|
+
result += self.diff_data(y.S2, x.S2)
|
|
1890
|
+
result += self.diff_data(y.S2L, x.S2L)
|
|
1891
|
+
else:
|
|
1892
|
+
result = self.diff_data(y.S0, x.S0, is_complex=False, sigma=sigma.S0)
|
|
1893
|
+
result += self.diff_data(y.S1, x.S1, sigma=sigma.S1)
|
|
1894
|
+
result += self.diff_data(y.P00, x.P00, sigma=sigma.P00)
|
|
1895
|
+
result += self.diff_data(y.S2, x.S2, sigma=sigma.S2)
|
|
1896
|
+
result += self.diff_data(y.S2L, x.S2L, sigma=sigma.S2L)
|
|
1897
|
+
|
|
1898
|
+
nval = (
|
|
1899
|
+
self.backend.bk_size(x.S0)
|
|
1900
|
+
+ self.backend.bk_size(x.P00)
|
|
1901
|
+
+ self.backend.bk_size(x.S1)
|
|
1902
|
+
+ self.backend.bk_size(x.S2)
|
|
1903
|
+
+ self.backend.bk_size(x.S2L)
|
|
1904
|
+
)
|
|
1905
|
+
|
|
1906
|
+
result /= self.backend.bk_cast(nval)
|
|
1907
|
+
else:
|
|
1908
|
+
return self.backend.bk_reduce_sum(x)
|
|
1909
|
+
return result
|
|
1326
1910
|
|
|
1327
|
-
def reduce_mean(self,x,axis=None):
|
|
1911
|
+
def reduce_mean(self, x, axis=None):
|
|
1328
1912
|
if axis is None:
|
|
1329
|
-
tmp=
|
|
1330
|
-
|
|
1331
|
-
|
|
1332
|
-
|
|
1333
|
-
|
|
1334
|
-
|
|
1335
|
-
|
|
1336
|
-
|
|
1337
|
-
|
|
1338
|
-
|
|
1339
|
-
|
|
1340
|
-
|
|
1913
|
+
tmp = (
|
|
1914
|
+
self.backend.bk_abs(self.backend.bk_reduce_sum(x.P00))
|
|
1915
|
+
+ self.backend.bk_abs(self.backend.bk_reduce_sum(x.S0))
|
|
1916
|
+
+ self.backend.bk_abs(self.backend.bk_reduce_sum(x.S1))
|
|
1917
|
+
+ self.backend.bk_abs(self.backend.bk_reduce_sum(x.S2))
|
|
1918
|
+
+ self.backend.bk_abs(self.backend.bk_reduce_sum(x.S2L))
|
|
1919
|
+
)
|
|
1920
|
+
|
|
1921
|
+
ntmp = (
|
|
1922
|
+
np.array(list(x.P00.shape)).prod()
|
|
1923
|
+
+ np.array(list(x.S0.shape)).prod()
|
|
1924
|
+
+ np.array(list(x.S1.shape)).prod()
|
|
1925
|
+
+ np.array(list(x.S2.shape)).prod()
|
|
1926
|
+
)
|
|
1927
|
+
|
|
1928
|
+
return tmp / ntmp
|
|
1341
1929
|
else:
|
|
1342
|
-
tmp=
|
|
1343
|
-
|
|
1344
|
-
|
|
1345
|
-
|
|
1346
|
-
|
|
1347
|
-
|
|
1348
|
-
|
|
1349
|
-
|
|
1350
|
-
|
|
1351
|
-
|
|
1352
|
-
|
|
1353
|
-
|
|
1354
|
-
|
|
1355
|
-
|
|
1356
|
-
|
|
1930
|
+
tmp = (
|
|
1931
|
+
self.backend.bk_abs(self.backend.bk_reduce_sum(x.P00, axis=axis))
|
|
1932
|
+
+ self.backend.bk_abs(self.backend.bk_reduce_sum(x.S0, axis=axis))
|
|
1933
|
+
+ self.backend.bk_abs(self.backend.bk_reduce_sum(x.S1, axis=axis))
|
|
1934
|
+
+ self.backend.bk_abs(self.backend.bk_reduce_sum(x.S2, axis=axis))
|
|
1935
|
+
+ self.backend.bk_abs(self.backend.bk_reduce_sum(x.S2L, axis=axis))
|
|
1936
|
+
)
|
|
1937
|
+
|
|
1938
|
+
ntmp = (
|
|
1939
|
+
np.array(list(x.P00.shape)).prod()
|
|
1940
|
+
+ np.array(list(x.S0.shape)).prod()
|
|
1941
|
+
+ np.array(list(x.S1.shape)).prod()
|
|
1942
|
+
+ np.array(list(x.S2.shape)).prod()
|
|
1943
|
+
+ np.array(list(x.S2L.shape)).prod()
|
|
1944
|
+
)
|
|
1945
|
+
|
|
1946
|
+
return tmp / ntmp
|
|
1947
|
+
|
|
1948
|
+
def reduce_sum(self, x, axis=None):
|
|
1357
1949
|
if axis is None:
|
|
1358
|
-
return
|
|
1359
|
-
self.backend.bk_reduce_sum(self.backend.bk_abs(x.
|
|
1360
|
-
self.backend.bk_reduce_sum(self.backend.bk_abs(x.
|
|
1361
|
-
self.backend.bk_reduce_sum(self.backend.bk_abs(x.
|
|
1362
|
-
self.backend.bk_reduce_sum(self.backend.bk_abs(x.
|
|
1950
|
+
return (
|
|
1951
|
+
self.backend.bk_reduce_sum(self.backend.bk_abs(x.P00))
|
|
1952
|
+
+ self.backend.bk_reduce_sum(self.backend.bk_abs(x.S0))
|
|
1953
|
+
+ self.backend.bk_reduce_sum(self.backend.bk_abs(x.S1))
|
|
1954
|
+
+ self.backend.bk_reduce_sum(self.backend.bk_abs(x.S2))
|
|
1955
|
+
+ self.backend.bk_reduce_sum(self.backend.bk_abs(x.S2L))
|
|
1956
|
+
)
|
|
1363
1957
|
else:
|
|
1364
|
-
return scat(
|
|
1365
|
-
|
|
1366
|
-
|
|
1367
|
-
|
|
1368
|
-
|
|
1369
|
-
|
|
1370
|
-
|
|
1371
|
-
|
|
1372
|
-
|
|
1373
|
-
|
|
1374
|
-
|
|
1375
|
-
|
|
1376
|
-
|
|
1377
|
-
|
|
1378
|
-
|
|
1379
|
-
|
|
1380
|
-
|
|
1381
|
-
|
|
1382
|
-
|
|
1383
|
-
|
|
1384
|
-
|
|
1385
|
-
|
|
1386
|
-
|
|
1387
|
-
|
|
1388
|
-
|
|
1389
|
-
|
|
1390
|
-
|
|
1958
|
+
return scat(
|
|
1959
|
+
self.backend.bk_reduce_sum(x.P00, axis=axis),
|
|
1960
|
+
self.backend.bk_reduce_sum(x.S0, axis=axis),
|
|
1961
|
+
self.backend.bk_reduce_sum(x.S1, axis=axis),
|
|
1962
|
+
self.backend.bk_reduce_sum(x.S2, axis=axis),
|
|
1963
|
+
self.backend.bk_reduce_sum(x.S2L, axis=axis),
|
|
1964
|
+
x.j1,
|
|
1965
|
+
x.j2,
|
|
1966
|
+
backend=self.backend,
|
|
1967
|
+
)
|
|
1968
|
+
|
|
1969
|
+
def ldiff(self, sig, x):
|
|
1970
|
+
return scat(
|
|
1971
|
+
x.domult(sig.P00, x.P00) * x.domult(sig.P00, x.P00),
|
|
1972
|
+
x.domult(sig.S0, x.S0) * x.domult(sig.S0, x.S0),
|
|
1973
|
+
x.domult(sig.S1, x.S1) * x.domult(sig.S1, x.S1),
|
|
1974
|
+
x.domult(sig.S2, x.S2) * x.domult(sig.S2, x.S2),
|
|
1975
|
+
x.domult(sig.S2L, x.S2L) * x.domult(sig.S2L, x.S2L),
|
|
1976
|
+
x.j1,
|
|
1977
|
+
x.j2,
|
|
1978
|
+
backend=self.backend,
|
|
1979
|
+
)
|
|
1980
|
+
|
|
1981
|
+
def log(self, x):
|
|
1982
|
+
return scat(
|
|
1983
|
+
self.backend.bk_log(x.P00),
|
|
1984
|
+
self.backend.bk_log(x.S0),
|
|
1985
|
+
self.backend.bk_log(x.S1),
|
|
1986
|
+
self.backend.bk_log(x.S2),
|
|
1987
|
+
self.backend.bk_log(x.S2L),
|
|
1988
|
+
x.j1,
|
|
1989
|
+
x.j2,
|
|
1990
|
+
backend=self.backend,
|
|
1991
|
+
)
|
|
1992
|
+
|
|
1993
|
+
def abs(self, x):
|
|
1994
|
+
return scat(
|
|
1995
|
+
self.backend.bk_abs(x.P00),
|
|
1996
|
+
self.backend.bk_abs(x.S0),
|
|
1997
|
+
self.backend.bk_abs(x.S1),
|
|
1998
|
+
self.backend.bk_abs(x.S2),
|
|
1999
|
+
self.backend.bk_abs(x.S2L),
|
|
2000
|
+
x.j1,
|
|
2001
|
+
x.j2,
|
|
2002
|
+
backend=self.backend,
|
|
2003
|
+
)
|
|
2004
|
+
|
|
2005
|
+
def inv(self, x):
|
|
2006
|
+
return scat(
|
|
2007
|
+
1 / (x.P00),
|
|
2008
|
+
1 / (x.S0),
|
|
2009
|
+
1 / (x.S1),
|
|
2010
|
+
1 / (x.S2),
|
|
2011
|
+
1 / (x.S2L),
|
|
2012
|
+
x.j1,
|
|
2013
|
+
x.j2,
|
|
2014
|
+
backend=self.backend,
|
|
2015
|
+
)
|
|
1391
2016
|
|
|
1392
2017
|
def one(self):
|
|
1393
|
-
return scat(1.0,1.0,1.0,1.0,1.0,[0],[0],backend=self.backend)
|
|
2018
|
+
return scat(1.0, 1.0, 1.0, 1.0, 1.0, [0], [0], backend=self.backend)
|
|
1394
2019
|
|
|
1395
|
-
@
|
|
1396
|
-
def eval_comp_fast(self, image1, image2=None,mask=None,Auto=True,s0_off=
|
|
2020
|
+
@tf_function
|
|
2021
|
+
def eval_comp_fast(self, image1, image2=None, mask=None, Auto=True, s0_off=1e-6):
|
|
1397
2022
|
|
|
1398
|
-
res=self.eval(image1, image2=image2,mask=mask,Auto=Auto,s0_off=s0_off)
|
|
1399
|
-
return res.P00,res.S0,res.S1,res.S2,res.S2L,res.j1,res.j2
|
|
2023
|
+
res = self.eval(image1, image2=image2, mask=mask, Auto=Auto, s0_off=s0_off)
|
|
2024
|
+
return res.P00, res.S0, res.S1, res.S2, res.S2L, res.j1, res.j2
|
|
1400
2025
|
|
|
1401
|
-
def eval_fast(self, image1, image2=None,mask=None,Auto=True,s0_off=
|
|
1402
|
-
p0,s0,s1,s2,s2l,j1,j2=self.eval_comp_fast(
|
|
1403
|
-
|
|
1404
|
-
|
|
1405
|
-
|
|
1406
|
-
|
|
2026
|
+
def eval_fast(self, image1, image2=None, mask=None, Auto=True, s0_off=1e-6):
|
|
2027
|
+
p0, s0, s1, s2, s2l, j1, j2 = self.eval_comp_fast(
|
|
2028
|
+
image1, image2=image2, mask=mask, Auto=Auto, s0_off=s0_off
|
|
2029
|
+
)
|
|
2030
|
+
return scat(p0, s0, s1, s2, s2l, j1, j2, backend=self.backend)
|