1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
|
__author__ = "Uwe Schmitt uschmitt@mineway.de, wrapped by V.A. Sole - ESRF"
__doc__ = """
This module is a simple wrapper to the py_nnma module of Uwe Schmitt (uschmitt@mineway.de)
in order to integrate it into PyMca. What follows is the documentation of py_nnma
py_nnma: python modules for nonnegative matrix approximation (NNMA)
(c) 2009 Uwe Schmitt, uschmitt@mineway.de
NNMA minimizes dist(Y, A X)
where: Y >= 0, m x n
A >= 0, m x k
X >= 0, n x k
k < min(m,n)
dist(A,B) can be || A - B ||_fro
or KL(A,B)
This moudule provides the following functions:
NMF, NMFKL, SNMF, RRI, ALS, GDCLS, GDCLS_L1, FNMAI, FNMAI_SPARSE,
NNSC and FastHALS
The common parameters when calling such a function are:
input:
Y -- the matrix for decomposition, maybe dense
from numpy or sparse from scipy.sparse
package
k -- number of componnets to estimate
Astart
Xstart -- matrices to start iterations. Maybe None
for using random start matrices.
eps -- termination swell value
maxcount -- max number of iterations to be performed
verbose -- if False: produce no output durint interations
if integer: give all 'verbose' itetations some
output about current state of iterations
output:
A, X -- result matrices of algorithm
obj -- value of objective function of last iteration
count -- number of iterations done
converged -- flag: indicates if iterations stoped within
max number of iterations
The following extra parameters exist depending on algorithm:
RRI : damping parameter 'psi' (default: 1e-12)
SNMF : sparsity parameter 'sparse_par' (default: 0)
ALS : regularization parameter 'regul' for stabilizing iterations
(default value 0). needed if objective value jitters.
GCDLS : 'regul' for l2-smoothness of X (default 0)
GDCLS_L1 : 'regul' for l1-smoothness of X (default 0)
FNMAI : 'stabil' for stabilizing algorithm (default value 1e-12)
'alpha' for stepsize (default value 0.1)
'tau' for number of inner iterations (default value 2)
FNMAI_SPARSE : as FNMAI plus
'regul' for l1-smoothness of X (default 0)
NNSC : 'alpha' for stepsize of gradient update of A
'sparse_par' for sparsity
This module is based on:
- Daniel D. Lee and H. Sebastian Seung:
"Algorithms for non-negative matrix factorization",
in Advances in Neural Information Processing 13
(Proc. NIPS*2000) MIT Press, 2001.
"Learning the parts of objects by non-negative matrix
factorization",
Nature, vol. 401, no. 6755, pp. 788-791, 1999.
- A. Cichocki and A-H. Phan:
"Fast local algorithms for large scale Nonnegative Matrix and
Tensor Factorizations",
IEICE Transaction on Fundamentals,
in print March 2009.
- P. O. Hoyer
"Non-negative Matrix Factorization with sparseness
constraints",
Journal of Machine Learning Research, vol. 5, pp. 1457-1469,
2004.
- Dongmin Kim, Suvrit Sra,Inderjit S. Dhillon:
"Fast Newton-type Methods for the Least Squares Nonnegative Matrix
Approximation Problem"
SIAM Data Mining (SDM), Apr. 2007
- Ngoc-Diep Ho:
dissertation from
http://edoc.bib.ucl.ac.be:81/ETD-db/collection/available/BelnUcetd-06052008-235205/
#############################################################################
Copyright (c) 2009 Uwe Schmitt, uschmitt@mineway.de
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
* Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials provided
* with the distribution. Neither the name of the <ORGANIZATION>
* nor the names of its contributors may be used to endorse or
* promote products derived from this software without specific
* prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
"""
import numpy
try:
import mdp
if mdp.__version__ >= '2.6':
MDP = True
else:
MDP = False
except:
MDP = False
from PyMca import py_nnma
DEBUG = 0
function_list = ['FNMAI', 'ALS', 'FastHALS', 'GDCLS']
function_dict = {"NNSC": py_nnma.NNSC,
"FNMAI_SPARSE": py_nnma.FNMAI_SPARSE,
"FNMAI": py_nnma.FNMAI,
"GDCLS_L1": py_nnma.GDCLS_L1,
"GDCLS": py_nnma.GDCLS,
"ALS": py_nnma.ALS,
"NMFKL": py_nnma.NMFKL,
"NMF": py_nnma.NMF,
"RRI": py_nnma.RRI,
"FastHALS": py_nnma.FastHALS,
"SNMF": py_nnma.SNMF,
}
def nnma(stack, ncomponents, binning=None,
function=None, eps=5e-5, verbose=DEBUG, maxcount=1000, kmeans=False):
if kmeans and (not MDP):
raise ValueError("K Means not supported")
#I take the defaults for the other parameters
param = dict(alpha=.1, tau=2, regul=1e-2, sparse_par=1e-1, psi=1e-3)
if function is None:
function = 'FNMAI'
nnma_function = function_dict[function]
if binning is None:
binning = 1
if hasattr(stack, "info") and hasattr(stack, "data"):
data = stack.data[:]
else:
data = stack[:]
oldShape = data.shape
if len(data.shape) == 3:
r, c, N = data.shape
if isinstance(data, numpy.ndarray):
data.shape = r*c, N
else:
r, N = data.shape
c = 1
if isinstance(data, numpy.ndarray):
if binning > 1:
data=numpy.reshape(data,[data.shape[0], data.shape[1]/binning, binning])
data=numpy.sum(data , axis=-1)
N=N/binning
else:
oldData = data
N = int(N/binning)
try:
data = numpy.zeros((r, c, N), oldData.dtype)
except MemoryError:
try:
data = numpy.zeros((r, c, N), numpy.float32)
except MemoryError:
text = "NNMAModule only works properly on numpy arrays.\n"
text += "Memory Error: Higher binning may help."
raise TypeError(text)
if binning == 1:
if len(oldShape) == 3:
for i in range(r):
data[i,:,:] = oldData[i,:,:]
data.shape = r * c, N
else:
data.shape = r * c, N
for i in range(r*c):
data[i,:] = oldData[i,:]
else:
if len(oldShape) == 3:
for i in range(r):
tmpData = oldData[i,:,:]
tmpData.shape = c, N, binning
data[i,:,:] = numpy.sum(tmpData, axis=-1)
data.shape = r * c, N
else:
data.shape = r * c, N
for i in range(r*c):
tmpData = oldData[i,:]
tmpData.shape = N, binning
data[i,:] = numpy.sum(tmpData, axis=-1)
#mindata = data.min()
#numpy.add(data, -mindata+1, data)
#I do not know the meaning of these paramenters
#py_nnma.scale(newdata)
param = dict(alpha=.1, tau=2, regul=1e-2, sparse_par=1e-1, psi=1e-3)
#Start tolerance
#1E+3 is conservative/fast
#1E-3 is probably slow
Astart = None
Xstart = None
#for i in range(start_ncomponents, ncomponents):
converged = False
while not converged:
A, X, obj, count, converged = nnma_function(data,
ncomponents,
Astart,
Xstart,
eps=eps,
maxcount=maxcount,
verbose=verbose,
**param)
if not converged:
print("WARNING: Possible problems converging")
#if binning > 1:
# numpy.add(data, mindata-1, data)
#data.shape = oldShape
images = A.T
if 0:
images.shape = ncomponents, r, c
return images, numpy.ones((ncomponents), numpy.float32),X
#order and scale images according to Gerd Wellenreuthers' recipe
#normalize all maps to be in the range [0, 1]
for i in range(ncomponents):
norm_factor = numpy.max(images[i, :])
if norm_factor > 0:
images[i, :] *= 1.0/norm_factor
X[i, :] *= norm_factor
#sort NNMA-spectra and maps
total_nnma_intensity = []
for i in range(ncomponents):
total_nnma_intensity += [[numpy.sum(images[i,:])*\
numpy.sum(X[i,:]), i]]
sorted_idx = [item[1] for item in sorted(total_nnma_intensity)]
sorted_idx.reverse()
#original data intensity
original_intensity = numpy.sum(data)
#final values
if kmeans:
n_more = 1
else:
n_more = 0
new_images = numpy.zeros((ncomponents + n_more, r*c), numpy.float32)
new_vectors = numpy.zeros((X.shape[0]+n_more, X.shape[1]), numpy.float32)
values = numpy.zeros((ncomponents+n_more,), numpy.float32)
for i in range(ncomponents):
idx = sorted_idx[i]
if 1:
new_images[i, :] = images[idx, :]
else:
#imaging the projected sum gives same results
Atmp = images[idx, :]
Atmp.shape = -r*c, 1
Xtmp = X[idx,:]
Xtmp.shape = 1, -1
new_images[i, :] = numpy.sum(numpy.dot(Atmp, Xtmp), axis=1)
new_vectors[i,:] = X[idx,:]
values[i] = 100.*total_nnma_intensity[idx][0]/original_intensity
new_images.shape = ncomponents + n_more, r, c
if kmeans:
classifier = mdp.nodes.KMeansClassifier(ncomponents)
for i in range(ncomponents):
classifier.train(new_vectors[i:i+1])
k = 0
for i in range(r):
for j in range(c):
spectrum = data[k:k+1,:]
new_images[-1, i,j] = classifier.label(spectrum)[0]
k += 1
return new_images, values, new_vectors
if __name__ == "__main__":
from PyMca import EDFStack
from PyMca import EdfFile
import os
import sys
import time
inputfile = "D:\DATA\COTTE\ch09\ch09__mca_0005_0000_0000.edf"
if len(sys.argv) > 1:
inputfile = sys.argv[1]
print(inputfile)
elif os.path.exists(inputfile):
print("Using a default test case")
else:
print("Usage:")
print("python NNMAModule.py indexed_edf_stack")
sys.exit(0)
stack = EDFStack.EDFStack(inputfile)
r0, c0, n0 = stack.data.shape
ncomponents = 10
outfile = os.path.basename(inputfile)+"ICA.edf"
e0 = time.time()
images, eigenvalues, eigenvectors = nnma(stack.data, ncomponents,
binning=1)
print("elapsed = %f" % (time.time() - e0))
if os.path.exists(outfile):
os.remove(outfile)
f = EdfFile.EdfFile(outfile)
for i in range(ncomponents):
f.WriteImage({}, images[i,:])
sys.exit(0)
|