-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmil_source_selection.py
441 lines (346 loc) · 15.9 KB
/
mil_source_selection.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
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
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Fri May 6 16:10:53 2022
@author: cmccurley
"""
"""
***********************************************************************
* File: mil_source_selection.py
* Name: Connor H. McCurley
* Date:
* Desc: Implementation of the Fusion-CAM algorithm by C. McCurley,
* "Discriminatve Feature Learning with Imprecise, Uncertain,
* and Ambiguous Data," Ph.D Thesis, Gainesville, FL, 2022.
*
* This code defines the source selection procedures.
*
*
* Author: Connor H. McCurley
* University of Florida, Electrical and Computer Engineering
* Email Address: [email protected]
* Latest Revision:
* This product is Copyright (c) 2022 University of Florida
* All rights reserved
**********************************************************************
"""
"""
%=====================================================================
%================= Import Settings and Packages ======================
%=====================================================================
"""
######################################################################
######################### Import Packages ############################
######################################################################
## General packages
import os
import json
import copy
from copy import deepcopy
import numpy as np
import itertools
from tqdm import tqdm
import dask.array as da
from scipy.spatial import distance_matrix
from sklearn.metrics.pairwise import cosine_similarity
import multiprocessing as mp
from skimage.filters import threshold_otsu
from sklearn.metrics import precision_recall_fscore_support as prfs
## Custom packages
from cm_MICI.util.cm_mi_choquet_integral_binary import BinaryMIChoquetIntegral
#from cm_MICI.util.cm_mi_choquet_integral_dask import MIChoquetIntegral
"""
%=====================================================================
%======================= Function Definitions ========================
%=====================================================================
"""
###############################################################################
############################# Dask MIL Selection ##############################
###############################################################################
def compute_fitness_binary_measure(indices, chi, Bags, Labels, NUM_SOURCES, Parameters):
tempBags = np.zeros((len(Bags)),dtype=np.object)
for idb, bag in enumerate(Bags):
tempbag = bag[:,indices]
tempBags[idb] = tempbag
## Train ChI with current set of sources
chi.train_chi(tempBags, Labels, Parameters)
## Return indices of NUM_SOURCES sources and corresponding fitness for the set of sources
return indices, chi.fitness
def select_mici_binary_measure(Bags, Labels, NUM_SOURCES, Parameters):
# print('\n Running selection for MICI Min-Max with Binary Measure... \n')
#######################################################################
########################### Select Sources ############################
#######################################################################
nAllSources = Bags[0].shape[1]
ind_set = []
fitness_set = []
# all_ind = np.array(list(itertools.combinations(range(nAllSources), NUM_SOURCES)))
all_ind = np.array(list(itertools.permutations(range(nAllSources), NUM_SOURCES)))
# all_measures = np.zeros((all_ind.shape[0],))
chi = BinaryMIChoquetIntegral()
#
# ## In parallel
# num_cores = mp.cpu_count()-2
# pool = mp.Pool(num_cores)
#
## res = [pool.apply_async(func=compute_fitness_binary_measure, args=(all_ind[k], chi, Bags, Labels, NUM_SOURCES, Parameters)) for k in tqdm(range(all_ind.shape[0]))]
#
# res = [pool.apply(func=compute_fitness_binary_measure, args=(all_ind[k], chi, Bags, Labels, NUM_SOURCES, Parameters)) for k in tqdm(range(all_ind.shape[0]))]
## res = [pool.apply(compute_fitness_binary_measure, (all_ind[k], chi, Bags, Labels, NUM_SOURCES, Parameters) for k in tqdm(range(all_ind.shape[0])))]
#
# print('Selected Sources!')
# print('Aggregating results...')
#
# with tqdm(total=all_ind.shape[0]) as progress_bar:
# for k in range(all_ind.shape[0]):
## ind_set.append(res[k].get()[0].tolist())
## fitness_set.append(res[k].get()[1])
# ind_set.append(res[k][0].tolist())
# fitness_set.append(res[k][1])
#
# if not(k % 20):
# progress_bar.update(20)
#
# pool.close()
# pool.join()
# ## Initialize selection with each potential source
# with tqdm(total=all_ind.shape[0]) as progress_bar:
# ## In series
# for k in range(all_ind.shape[0]):
#
# ind, fitness = compute_fitness_binary_measure(all_ind[k], chi, Bags, Labels, NUM_SOURCES, Parameters)
#
# ind_set.append(ind.tolist())
# fitness_set.append(fitness)
#
# progress_bar.update()
# search = True
## Initial search to remove sources
sources_to_remove = []
sources_to_keep = []
with tqdm(total=Parameters.initial_num_sources) as progress_bar:
for idx in range(Parameters.initial_num_sources):
if idx == 7:
print('here')
ind = np.where(all_ind[:,0] == idx)
tempBags = np.zeros((len(Bags)),dtype=np.object)
for idb, bag in enumerate(Bags):
tempbag = bag[:,all_ind[ind[0][0],:]]
tempBags[idb] = tempbag
## Train ChI with current set of sources
chi.train_chi(tempBags, Labels, Parameters)
if (chi.measure[0] == 0):
all_ind = np.delete(all_ind, ind, 0)
sources_to_remove.append(idx)
else:
sources_to_keep.append(idx)
progress_bar.update()
## Genearte new sources to search
all_ind = np.array(list(itertools.combinations(range(nAllSources), NUM_SOURCES)))
sources = []
for idx in sources_to_keep:
ind = np.where(all_ind == idx)[0]
sources.append(all_ind[ind,:])
if len(sources_to_keep) > 1:
all_ind = np.vstack(sources)
## In parallel
num_cores = mp.cpu_count()-2
pool = mp.Pool(num_cores)
res = [pool.apply(func=compute_fitness_binary_measure, args=(all_ind[k], chi, Bags, Labels, NUM_SOURCES, Parameters)) for k in tqdm(range(all_ind.shape[0]))]
print('Selected Sources!')
print('Aggregating results...')
with tqdm(total=all_ind.shape[0]) as progress_bar:
for k in range(all_ind.shape[0]):
ind_set.append(res[k][0].tolist())
fitness_set.append(res[k][1])
if not(k % 20):
progress_bar.update(20)
pool.close()
pool.join()
print('Done! Returning best sources.')
#######################################################################
########################### Save Results ##############################
#######################################################################
## Order fitness values from greatest to least
top_order = (-np.array(fitness_set)).argsort().astype(np.int16).tolist()
top_order_ind = int(top_order[0])
## Values to return - indices of selected sources
return_indices = ind_set[top_order_ind]
return return_indices
###############################################################################
############################# Dask MIL Selection ##############################
###############################################################################
def compute_iou_binary_measure(indices, chi, Bags, Labels, NUM_SOURCES, Bags_test, Labels_test, Parameters):
tempBags = np.zeros((len(Bags)),dtype=np.object)
for idb, bag in enumerate(Bags):
tempbag = bag[:,indices]
tempBags[idb] = tempbag
## Train ChI with current set of sources
chi.train_chi(tempBags, Labels, Parameters)
## Compute output with mean measure elements
iou = np.zeros((len(Bags_test)))
mse = np.zeros((len(Bags_test)))
metric_precision = np.zeros((len(Bags_test)))
metric_recall = np.zeros((len(Bags_test)))
metric_f1_score = np.zeros((len(Bags_test)))
if (Parameters.n_choose_k == 'y'):
## Train ChI with selected sources
tempBags = np.zeros((len(Bags_test),),dtype=np.object)
for idb, bag in enumerate(Bags_test):
temp_bag = bag[:,indices]
tempBags[idb] = temp_bag
currentBagsTest = deepcopy(tempBags)
else:
currentBagsTest = deepcopy(Bags_test)
for idx in range(len(currentBagsTest)):
y_true = Labels_test[idx].reshape((Labels_test[idx].shape[0]*Labels_test[idx].shape[1])).astype('float64')
tempTestBag = np.zeros((1,),dtype=np.object)
tempTestBag[0] = currentBagsTest[idx]
y_est = chi.compute_chi(tempTestBag,len(tempTestBag),chi.measure)
error = y_true - y_est
mse[idx] = np.dot(error,error)/(2*len(y_true))
## Compute iou
gt_img = y_true > 0
try:
if(Parameters.CAM_SEG_THRESH == 0):
img_thresh = threshold_otsu(y_est)
else:
img_thresh = Parameters.CAM_SEG_THRESH
binary_feature_map = y_est > img_thresh
except:
binary_feature_map = y_est > 0.1
## Compute fitness as IoU to pseudo-groundtruth
intersection = np.logical_and(binary_feature_map, gt_img)
union = np.logical_or(binary_feature_map, gt_img)
try:
iou[idx] = np.sum(intersection) / np.sum(union)
except:
iou[idx] = 0
prec, rec, f1, _ = prfs(gt_img, binary_feature_map, pos_label=1, average='binary')
metric_precision[idx], metric_recall[idx], metric_f1_score[idx] = round(prec,5), round(rec,5), round(f1,5)
results_miou_mean = np.mean(iou)
results_miou_std = np.std(iou)
results_mprec_mean = np.mean(metric_precision)
results_mprec_std = np.std(metric_precision)
results_mrec_mean = np.mean(metric_recall)
results_mrec_std = np.std(metric_recall)
results_mf1_mean = np.mean(metric_f1_score)
results_mf1_std = np.std(metric_f1_score)
## Return indices of NUM_SOURCES sources and corresponding fitness for the set of sources
return indices, chi.fitness, results_miou_mean, results_mprec_mean, results_mrec_mean, results_mf1_mean, results_miou_std, results_mprec_std, results_mrec_std, results_mf1_std
def select_mici_binary_measure_validation_iou(Bags, Labels, NUM_SOURCES, Bags_test, Labels_test, Parameters):
# print('\n Running selection for MICI Min-Max with Binary Measure... \n')
#######################################################################
########################### Select Sources ############################
#######################################################################
nAllSources = Bags[0].shape[1]
ind_set = []
fitness_set = []
iou_set = []
precision_set = []
recall_set = []
f1_set = []
iou_set_std = []
precision_set_std = []
recall_set_std = []
f1_set_std = []
# all_ind = np.array(list(itertools.combinations(range(nAllSources), NUM_SOURCES)))
all_ind = np.array(list(itertools.permutations(range(nAllSources), NUM_SOURCES)))
# all_measures = np.zeros((all_ind.shape[0],))
chi = BinaryMIChoquetIntegral()
# ## In parallel
# num_cores = mp.cpu_count()-2
# pool = mp.Pool(num_cores)
#
## res = [pool.apply_async(func=compute_fitness_binary_measure, args=(all_ind[k], chi, Bags, Labels, NUM_SOURCES, Parameters)) for k in tqdm(range(all_ind.shape[0]))]
#
# res = [pool.apply(func=compute_iou_binary_measure, args=(all_ind[k], chi, Bags, Labels, NUM_SOURCES, Bags_test, Labels_test, Parameters)) for k in tqdm(range(all_ind.shape[0]))]
## res = [pool.apply(compute_fitness_binary_measure, (all_ind[k], chi, Bags, Labels, NUM_SOURCES, Parameters) for k in tqdm(range(all_ind.shape[0])))]
#
# print('Selected Sources!')
# print('Aggregating results...')
#
# with tqdm(total=all_ind.shape[0]) as progress_bar:
# for k in range(all_ind.shape[0]):
## ind_set.append(res[k].get()[0].tolist())
## fitness_set.append(res[k].get()[1])
# ind_set.append(res[k][0].tolist())
# fitness_set.append(res[k][1])
# iou_set.append(res[k][2])
# precision_set.append(res[k][3])
# recall_set.append(res[k][4])
# f1_set.append(res[k][5])
#
# if not(k % 20):
# progress_bar.update(20)
#
# pool.close()
# pool.join()
# ## Initialize selection with each potential source
# with tqdm(total=all_ind.shape[0]) as progress_bar:
# ## In series
# for k in range(all_ind.shape[0]):
#
# ind, fitness = compute_fitness_binary_measure(all_ind[k], chi, Bags, Labels, NUM_SOURCES, Parameters)
#
# ind_set.append(ind.tolist())
# fitness_set.append(fitness)
#
# progress_bar.update()
#
# progress_bar.reset()
## Initial search to remove sources
sources_to_remove = []
sources_to_keep = []
with tqdm(total=Parameters.initial_num_sources) as progress_bar:
for idx in range(Parameters.initial_num_sources):
if idx == 7:
print('here')
ind = np.where(all_ind[:,0] == idx)
tempBags = np.zeros((len(Bags)),dtype=np.object)
for idb, bag in enumerate(Bags):
tempbag = bag[:,all_ind[ind[0][0],:]]
tempBags[idb] = tempbag
## Train ChI with current set of sources
chi.train_chi(tempBags, Labels, Parameters)
if (chi.measure[0] == 0):
all_ind = np.delete(all_ind, ind, 0)
sources_to_remove.append(idx)
else:
sources_to_keep.append(idx)
progress_bar.update()
## Genearte new sources to search
all_ind = np.array(list(itertools.combinations(range(nAllSources), NUM_SOURCES)))
sources = []
for idx in sources_to_keep:
ind = np.where(all_ind == idx)[0]
sources.append(all_ind[ind,:])
if len(sources_to_keep) > 1:
all_ind = np.vstack(sources)
## In parallel
num_cores = mp.cpu_count()-2
pool = mp.Pool(num_cores)
res = [pool.apply(func=compute_iou_binary_measure, args=(all_ind[k], chi, Bags, Labels, NUM_SOURCES, Bags_test, Labels_test, Parameters)) for k in tqdm(range(all_ind.shape[0]))]
print('Selected Sources!')
print('Aggregating results...')
with tqdm(total=all_ind.shape[0]) as progress_bar:
for k in range(all_ind.shape[0]):
ind_set.append(res[k][0].tolist())
fitness_set.append(res[k][1])
iou_set.append(res[k][2])
precision_set.append(res[k][3])
recall_set.append(res[k][4])
f1_set.append(res[k][5])
iou_set_std.append(res[k][6])
precision_set_std.append(res[k][7])
recall_set_std.append(res[k][8])
f1_set_std.append(res[k][9])
if not(k % 20):
progress_bar.update(20)
pool.close()
pool.join()
print('Done! Returning best sources.')
#######################################################################
########################### Save Results ##############################
#######################################################################
return ind_set, fitness_set, iou_set, precision_set, recall_set, f1_set, iou_set_std, precision_set_std, recall_set_std, f1_set_std