-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpredict_batch.py
998 lines (886 loc) · 56.5 KB
/
predict_batch.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
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
import os
import shutil
import sys
import time
from datetime import datetime
from typing import Union
import matplotlib.pyplot as plt
import numpy as np
import torch
from torch.utils.data import DataLoader
import hardware
import loader
import mil_metrics
import models
import r
from util import data_renderer
from util import log
from util import paths
from util import utils
from util.omnisphero_data_loader import OmniSpheroDataLoader
from util.paths import debug_prediction_dirs_win
from util.paths import default_out_dir_unix_base
from util.utils import line_print
from util.well_metadata import TileMetadata
from util.well_metadata import extract_well_info
def predict_path(model_save_path: str, checkpoint_file: str, bag_paths: [str], normalize_enum: int, out_dir: str,
max_workers: int, image_folder: str, channel_inclusions=loader.default_channel_inclusions_all,
tile_constraints=loader.default_tile_constraints_nuclei, global_log_dir: str = None,
sigmoid_verbose: bool = False,
skip_already_predicted: bool = False,
render_attention_spheres_enabled: bool = True,
render_attention_instance_range_min: float = None,
render_attention_instance_range_max: float = None,
render_dose_response_curves_enabled: bool = True, hist_bins_override=None,
render_merged_predicted_tiles_activation_overlays: bool = False, gpu_enabled: bool = False,
render_attention_cell_distributions: bool = False,
render_attention_histogram_enabled: bool = False,
render_attention_cytometry_prediction_distributions_enabled: bool = False,
oligo_positive_z_score_scale: float = 1.0, oligo_z_score_max_kernel_size: int = 1,
predict_samples_as_bags: bool = False,
data_loader_data_saver: bool = False,
used_tile_quartiles=None,
clear_global_logs: bool = True,
out_image_dpi: int = 300,
):
start_time = datetime.now()
log_label = str(start_time.strftime("%d-%m-%Y-%H-%M-%S"))
if used_tile_quartiles is None:
used_tile_quartiles = loader.default_used_tile_quartiles.copy()
used_tile_quartiles_enum = utils.boolean_to_integer(used_tile_quartiles[0],
used_tile_quartiles[1],
used_tile_quartiles[2],
used_tile_quartiles[3])
# Setting up log
global_log_filename = None
local_log_filename = out_dir + os.sep + 'log_predictions.txt'
print('Local Log file: ' + str(local_log_filename))
log.add_file(local_log_filename)
if global_log_dir is not None:
global_log_filename = global_log_dir + os.sep + 'log-predictions-' + log_label + '.txt'
os.makedirs(global_log_dir, exist_ok=True)
log.add_file(global_log_filename)
log.diagnose()
# Checking if already exists!
log.write('Saving predictions here: ' + out_dir)
out_dir_exists = os.path.exists(out_dir)
log.write('Does this path already exist? -> ' + str(out_dir_exists))
log.write('Skipping already predicted enabled? -> ' + str(skip_already_predicted))
if out_dir_exists and skip_already_predicted:
log.write('That path already exists and existing predictions should be skipped.')
time.sleep(0.42)
log.write('#################')
log.write('### Skipping! ###')
log.write('#################')
print('\n')
time.sleep(1.1337)
return
else:
log.write('Overwriting existing predictions (if there are any!)')
time.sleep(2)
# Setting up dirs & paths
log.write('Predicting from paths: ' + str(bag_paths))
if not model_save_path.endswith(".h5"):
model_save_path = model_save_path + "model.h5"
model_state_dict_file = model_save_path[:-8] + 'model.pt'
os.makedirs(out_dir, exist_ok=True)
log.write('Predicting to: ' + out_dir)
# Setting the device calculations will take part in
device = hardware.get_hardware_device(gpu_preferred=gpu_enabled)
log.write('Prediction device: ' + str(device))
time.sleep(2)
log.write('z-score max kernel size: ' + str(oligo_z_score_max_kernel_size))
log.write('z-score positive scale: ' + str(oligo_positive_z_score_scale))
time.sleep(5)
# Loading model
log.write('Loading model: ' + model_save_path)
log.write('Assumed model state dict: ' + model_state_dict_file)
model = None
loading_error = False
if os.path.exists(model_save_path):
try:
model = torch.load(model_save_path, map_location='cpu')
if not isinstance(model, models.OmniSpheroMil):
log.write('Warning! Loaded data is the wrong type! Loaded Type: ' + str(type(model)))
loading_error = True
except Exception as e:
log.write("Error loading model!")
log.write(str(e))
loading_error = True
if (model is None or loading_error) and os.path.exists(model_state_dict_file):
loss_function = 'binary_cross_entropy'
accuracy_function = 'binary'
log.write('Creating a new untrained model and feeding it the state dict')
log.write('Selected device: ' + str(device))
model = models.BaselineMIL((3, 150, 150), device=device, loss_function=None, accuracy_function=None)
model.load_state_dict(torch.load(model_state_dict_file, map_location='cpu'))
elif model is None or loading_error or not os.path.exists(model_state_dict_file):
log.write('Model is None: ' + str(model is None))
log.write('Has loading error: ' + str(loading_error))
log.write('Exists state dict: ' + str(os.path.exists(model_state_dict_file)))
log.write("\n\n ==== WARNING! INCORRECT MODEL PATH. YOUR MODEL MAY NOT HAVE BEEN LOADED ====\n\n")
# Checking if the right data has been loaded
assert model is not None
assert isinstance(model, models.OmniSpheroMil)
model = model.eval()
# Loading best checkpoint
model, _, _, _ = models.load_checkpoint(load_path=checkpoint_file, model=model, optimizer=None, map_location='cpu')
log.write('Finished loading the model.')
# Updating the model device
model.device = device
time.sleep(0.69)
# Notifying about the quartiles to be predicted
log.write('Using neurosphere tile quartiles: ' + str(used_tile_quartiles_enum) + ' (Enum)')
log.write('Using neurosphere tile quartiles: ' + str(used_tile_quartiles))
time.sleep(1)
# Loading the data
X, y, y_tiles, X_raw, X_metadata, bag_names, experiment_names, _, well_names, errors, loaded_files_list = loader.load_bags_json_batch(
batch_dirs=bag_paths,
max_workers=max_workers,
include_raw=True,
force_balanced_batch=False,
channel_inclusions=channel_inclusions,
used_tile_quartiles=used_tile_quartiles,
constraints_0=tile_constraints,
constraints_1=tile_constraints,
label_0_well_indices=loader.default_well_indices_all,
label_1_well_indices=loader.default_well_indices_all,
positive_z_score_scale=oligo_positive_z_score_scale,
z_score_max_kernel_size=oligo_z_score_max_kernel_size,
normalize_enum=normalize_enum)
X = [np.einsum('bhwc->bchw', bag) for bag in X]
# Printing the size in memory
X_s = str(utils.byteSizeString(utils.listToBytes(X)))
X_s_raw = str(utils.byteSizeString(utils.listToBytes(X_raw)))
y_s = str(utils.byteSizeString(sys.getsizeof(y)))
log.write("X-size in memory (after loading all data): " + str(X_s))
log.write("y-size in memory (after loading all data): " + str(y_s))
log.write("X-size (raw) in memory (after loading all data): " + str(X_s_raw))
del X_s, y_s, X_s_raw
dataset, input_dim = loader.convert_bag_to_batch(bags=X, labels=None, y_tiles=None)
data_loader = DataLoader(dataset, batch_size=1, shuffle=False,
pin_memory=False,
num_workers=max_workers)
del X, dataset, y, y_tiles
# TODO react to loading errors
log.write('Number of files loaded: ' + str(len(loaded_files_list)))
log.write('Number of loading errors: ' + str(len(errors)))
if len(errors) > 1:
log.write("[!!]\n[ERRORS WHILE LOADING EXPERIMENTS!]\n[!!]", include_timestamp=False)
for i in range(len(errors)):
e = errors[i]
log.write('Error #' + str(i) + ': "' + str(e) + '".')
log.write_exception(e)
del e
del i
del errors, loaded_files_list
norm = True
sparse = True
predict_data(model=model, data_loader=data_loader, X_raw=X_raw, X_metadata=X_metadata,
experiment_names=experiment_names, input_dim=input_dim, sparse_hist=sparse,
hist_bins_override=hist_bins_override, out_image_dpi=out_image_dpi,
normalized_attention=norm, clear_old_data=False, image_folder=image_folder,
sigmoid_verbose=sigmoid_verbose, render_attention_histogram_enabled=render_attention_histogram_enabled,
render_merged_predicted_tiles_activation_overlays=render_merged_predicted_tiles_activation_overlays,
render_attention_spheres_enabled=render_attention_spheres_enabled,
render_dose_response_curves_enabled=render_dose_response_curves_enabled,
render_attention_cell_distributions=render_attention_cell_distributions,
render_attention_instance_range_min=render_attention_instance_range_min,
render_attention_instance_range_max=render_attention_instance_range_max,
predict_samples_as_bags=predict_samples_as_bags,
num_workers=max_workers,
render_attention_cytometry_prediction_distributions_enabled=render_attention_cytometry_prediction_distributions_enabled,
well_names=well_names, out_dir=out_dir)
del data_loader, X_raw, X_metadata
# All done and finishing up the logger
log.write('Batch prediction finished.')
log.remove_file(local_log_filename)
if global_log_dir is not None:
log.remove_file(global_log_filename)
if clear_global_logs:
log.clear_files()
def _predict_samples_as_bags(model, data_loader, X_raw: [np.ndarray], num_workers: int,
X_metadata: [TileMetadata], experiment_names: [str], well_names: [str], image_folder: str,
input_dim: (int), out_dir: str, dpi: 500):
# TODO update arguments
data_set = data_loader.dataset
# Making sure all's well that predicts well
assert len(X_raw) == len(X_metadata)
assert len(experiment_names) == len(X_metadata)
assert len(well_names) == len(X_metadata)
assert len(well_names) == len(data_set)
log.write('Predicting the samples of very bag as a bag.')
for i in range(len(X_metadata)):
dataset_current = data_set[i]
X_metadata_current = X_metadata[i]
X_raw_current = X_raw[i]
experiment_name_current = experiment_names[i]
well_name_current = well_names[i]
sample_count = X_raw_current.shape[0]
assert len(X_metadata_current) == sample_count
assert len(X_metadata_current) == dataset_current[0].shape[0]
metadata = X_metadata_current[0]
log.write('[' + str(i + 1) + '/' + str(
len(X_metadata)) + '] Predicting sample as bag for: ' + experiment_name_current + ' - ' + well_name_current)
# iterating over every sample
y_hat_samples = []
out_image_localized_positive_raw = np.zeros((metadata.well_image_height, metadata.well_image_width, 3),
dtype=np.uint8)
out_image_localized_overlay = np.zeros((metadata.well_image_height, metadata.well_image_width, 3),
dtype=np.uint8)
out_image_localized_overlay_map_r = np.ones((metadata.well_image_height, metadata.well_image_width),
dtype=np.uint8)
out_image_localized_overlay_map_g = np.ones((metadata.well_image_height, metadata.well_image_width),
dtype=np.uint8)
out_image_localized_overlay_map_b = np.ones((metadata.well_image_height, metadata.well_image_width),
dtype=np.uint8)
out_image_localized_overlay_map_mask = np.zeros((metadata.well_image_height, metadata.well_image_width),
dtype=np.uint8)
positive_samples = []
positive_sample_count = 0
task_time_durations = []
for j in range(sample_count):
task_start_time = datetime.now()
log.write('[' + str(i + 1) + '/' + str(
len(X_metadata)) + '] Predicting sample as bag for: ' + experiment_name_current + ' - ' + well_name_current + ': ' + str(
j + 1) + '/' + str(sample_count), include_in_files=False)
X_raw_sample = X_raw_current[j]
X_metadata_sample = X_metadata_current[j]
X_sample = dataset_current[0][j]
# Expanding sample so it can be used for predictions
X_sample = X_sample.copy()
X_sample = np.copy(X_sample)
X_sample = np.expand_dims(X_sample, axis=0)
data_set_sample, _ = loader.convert_bag_to_batch(bags=[X_sample], labels=None, y_tiles=None)
data_loader = DataLoader(data_set_sample, batch_size=1, shuffle=False,
pin_memory=False,
num_workers=num_workers)
y_hats, y_preds, _, _, y_samples_pred, _, all_attentions, original_bag_indices = models.get_predictions(
model, data_loader)
y_pred = bool(y_preds[0])
del y_hats, y_samples_pred, all_attentions, original_bag_indices, y_preds
# Saving the results
y_hat_samples.append(y_pred)
X_raw_sample = np.copy(X_raw_sample)
X_raw_sample = X_raw_sample.copy()
width, height, _ = X_raw_sample.shape
pos_x: int = int(X_metadata_sample.pos_x)
pos_y: int = int(X_metadata_sample.pos_y)
out_image_localized_overlay[pos_y:pos_y + height, pos_x:pos_x + width] = X_raw_sample
if y_pred:
positive_sample_count = positive_sample_count + 1
positive_samples.append(X_raw_sample)
out_image_localized_positive_raw[pos_y:pos_y + height, pos_x:pos_x + width] = X_raw_sample
out_image_localized_overlay_map_r[pos_y:pos_y + height, pos_x:pos_x + width] = 1
out_image_localized_overlay_map_b[pos_y:pos_y + height, pos_x:pos_x + width] = 0
out_image_localized_overlay_map_g[pos_y:pos_y + height, pos_x:pos_x + width] = 0
out_image_localized_overlay_map_mask[pos_y:pos_y + height, pos_x:pos_x + width] = 1
out_image_localized_overlay_rgb = out_image_localized_overlay[pos_y:pos_y + height, pos_x:pos_x + width]
out_image_localized_overlay_r = out_image_localized_overlay_rgb[:, :, 0] * 1.25
out_image_localized_overlay_g = out_image_localized_overlay_rgb[:, :, 1] * 0.75
out_image_localized_overlay_b = out_image_localized_overlay_rgb[:, :, 2] * 0.75
out_image_localized_overlay_r = out_image_localized_overlay_r * 1.0
out_image_localized_overlay_r = out_image_localized_overlay_r.astype(np.uint8)
out_image_localized_overlay_g = out_image_localized_overlay_g.astype(np.uint8)
out_image_localized_overlay_b = out_image_localized_overlay_b.astype(np.uint8)
out_image_localized_overlay_rgb = np.dstack(
(out_image_localized_overlay_r, out_image_localized_overlay_g, out_image_localized_overlay_b))
out_image_localized_overlay_rgb = out_image_localized_overlay_rgb.astype('uint8')
task_end_time = datetime.now()
task_time_diff = task_end_time - task_start_time
task_time_durations.append(task_time_diff)
mean_task_time = np.mean(task_time_durations)
current_time = datetime.now()
eta_time = current_time + (mean_task_time * (sample_count - (j + 1)))
log.write('Task finished ETA: ' + str(eta_time.strftime("%d/%m/%Y, %H:%M:%S")), include_in_files=False,
include_timestamp=False)
log.write('FINISHED ALL SAMPLES FOR THIS BAG!')
log.write('Number of positive bags: ' + str(positive_sample_count))
out_dir_current = out_dir + os.sep + experiment_name_current + os.sep + well_name_current + os.sep
os.makedirs(out_dir_current, exist_ok=True)
log.write('Saving to: ' + out_dir_current)
# Updating the masks
out_image_localized_overlay_map_r = out_image_localized_overlay_map_r * 255
out_image_localized_overlay_map_g = out_image_localized_overlay_map_g * 255
out_image_localized_overlay_map_b = out_image_localized_overlay_map_b * 255
out_image_localized_overlay_r = out_image_localized_overlay[:, :, 0]
out_image_localized_overlay_g = out_image_localized_overlay[:, :, 1]
out_image_localized_overlay_b = out_image_localized_overlay[:, :, 2]
out_image_localized_overlay_r = out_image_localized_overlay_map_r * 0.5 + out_image_localized_overlay_r * 0.5
out_image_localized_overlay_g = out_image_localized_overlay_map_g * 0.5 + out_image_localized_overlay_g * 0.5
out_image_localized_overlay_b = out_image_localized_overlay_map_b * 0.5 + out_image_localized_overlay_b * 0.5
out_image_localized_overlay_map_mask = out_image_localized_overlay_map_mask.astype(np.bool)
out_image_localized_overlay[:, :, 0][out_image_localized_overlay_map_mask] = out_image_localized_overlay_r[
out_image_localized_overlay_map_mask]
out_image_localized_overlay[:, :, 1][out_image_localized_overlay_map_mask] = out_image_localized_overlay_g[
out_image_localized_overlay_map_mask]
out_image_localized_overlay[:, :, 2][out_image_localized_overlay_map_mask] = out_image_localized_overlay_b[
out_image_localized_overlay_map_mask]
# TODO move all this rendering to the data renderer file
# Rendering the data
if positive_sample_count > 0:
for j in range(len(positive_samples)):
sample = positive_samples[j]
# detail_image_name = out_dir_current_detail + os.sep + current_experiment_name + '-' + current_well + '-' + str(
# j) + '.png'
# plt.imsave(detail_image_name, sample)
positive_samples[j] = mil_metrics.outline_rgb_array(sample, None, None, bright_mode=True,
override_colormap=[255, 255, 255])
matching_samples_fused = mil_metrics.fuse_image_tiles(images=positive_samples, light_mode=False)
fused_image_name = out_dir_current + os.sep + experiment_name_current + '-' + well_name_current + '_fused.png'
fused_image_name_detail = out_dir_current + os.sep + experiment_name_current + '-' + well_name_current + '_fused-detail.png'
plt.imsave(fused_image_name, matching_samples_fused)
localized_image_name = out_dir_current + os.sep + experiment_name_current + '-' + well_name_current + '_localized.png'
localized_image_name_overlay = out_dir_current + os.sep + experiment_name_current + '-' + well_name_current + '_localized-overlay.png'
plt.imsave(localized_image_name, out_image_localized_positive_raw)
plt.imsave(localized_image_name_overlay, out_image_localized_overlay)
plt.clf()
plt.imshow(out_image_localized_overlay)
plt.title(experiment_name_current + ' - ' + well_name_current + '\nPositive Samples: ' + str(
positive_sample_count) + '/' + str(sample_count))
plt.xticks([])
plt.yticks([])
plt.tight_layout()
plt.autoscale()
plt.savefig(fused_image_name_detail + '.png', dpi=dpi)
plt.savefig(fused_image_name_detail + '.svg', dpi=dpi, transparent=True)
plt.savefig(fused_image_name_detail + '.pdf', dpi=dpi)
log.write('Done: Predicting the samples of very bag as a bag.')
def predict_data(model: models.BaselineMIL, data_loader: Union[OmniSpheroDataLoader, DataLoader], X_raw: [np.ndarray],
X_metadata: [TileMetadata], experiment_names: [str], well_names: [str], image_folder: str,
num_workers: int,
input_dim: (int), out_dir: str, sparse_hist: bool = True, hist_bins_override=None,
normalized_attention: bool = True, save_sigmoid_plot: bool = True, sigmoid_verbose: bool = False,
render_dose_response_curves_enabled: bool = True, histogram_bootstrap_replications: int = 1000,
histogram_resample_size: int = 100, attention_normalized_metrics: bool = True,
render_attention_spheres_enabled: bool = False, render_attention_histogram_enabled: bool = False,
render_attention_cytometry_prediction_distributions_enabled: bool = False,
render_attention_instance_range_min: float = None,
render_attention_instance_range_max: float = None,
predict_samples_as_bags: bool = False,
render_merged_predicted_tiles_activation_overlays: bool = False, clear_old_data: bool = False,
render_attention_cell_distributions: bool = False, out_image_dpi: int = 250):
os.makedirs(out_dir, exist_ok=True)
if not type(render_attention_instance_range_min) == type(render_attention_instance_range_max):
if not (type(render_attention_instance_range_min) == type(None) or type(
render_attention_instance_range_min) == type(float) or type(
render_attention_instance_range_max) == type(None) or type(render_attention_instance_range_max) == type(
float)):
assert False
log.write('Using image folder: ' + image_folder)
log.write('Exists image folder: ' + str(os.path.exists(image_folder)))
log.write('R - Is pyRserve connection available: ' + str(r.has_connection(also_test_script=True)))
log.write('Predicting ' + str(len(X_metadata)) + ' bags.')
log.write('Saving predictions to: ' + out_dir)
y_hats, y_preds, _, _, y_samples_pred, _, all_attentions, original_bag_indices = models.get_predictions(
model, data_loader)
assert len(X_metadata) == len(all_attentions)
assert len(X_metadata) == len(X_raw)
#####################################################
# PREDICTING SAMPLES AS IF THEY WERE BAGS
#####################################################
if predict_samples_as_bags:
out_dir_samples_as_bags = out_dir + os.sep + 'sample_predictions'
os.makedirs(out_dir_samples_as_bags, exist_ok=True)
_predict_samples_as_bags(model=model, data_loader=data_loader, X_raw=X_raw, X_metadata=X_metadata,
experiment_names=experiment_names, well_names=well_names, image_folder=image_folder,
num_workers=num_workers, input_dim=input_dim, dpi=out_image_dpi,
out_dir=out_dir_samples_as_bags)
del out_dir_samples_as_bags
del data_loader, model
#####################################################
# APPLYING THE MODEL, DOING THE PREDICTIONS
#####################################################
# Evaluating attention metrics (histogram, entropy, etc.)
attention_metadata_list, attention_n_list, attention_bins_list, attention_otsu_index_list, attention_otsu_threshold_list, attention_entropy_attention_list, attention_entropy_hist_list, error_list = mil_metrics.attention_metrics_batch(
all_attentions=all_attentions,
X_metadata=X_metadata,
hist_bins_override=hist_bins_override,
normalized=attention_normalized_metrics)
# WRITING ERRORS (if they exist)
log.write(" === START ERROR LIST ===")
for error in error_list:
log.write("ERROR WHILE PREDICTING: " + str(error))
log.write_exception(error)
log.write(" === END ERROR LIST ===")
#####################################################
# RENDERING SAMPLES IN A GIVEN ATTENTION INTERVAL
#####################################################
if render_attention_instance_range_min is not None and render_attention_instance_range_max is not None:
render_attention_instance_range_min = float(render_attention_instance_range_min)
render_attention_instance_range_max = float(render_attention_instance_range_max)
assert render_attention_instance_range_min <= render_attention_instance_range_max
out_dir_attention_range = out_dir + os.sep + 'significant-attentions-' + str(
render_attention_instance_range_min) + '-' + str(render_attention_instance_range_max) + os.sep
os.makedirs(out_dir_attention_range, exist_ok=True)
data_renderer.render_attention_instance_range(out_dir=out_dir_attention_range,
X_metadata=X_metadata,
y_preds=y_preds,
all_attentions=all_attentions,
X_raw=X_raw,
render_attention_instance_range_min=render_attention_instance_range_min,
render_attention_instance_range_max=render_attention_instance_range_max
)
del out_dir_attention_range
#####################################################
# RENDERING CYTOMETRY PAPER PREDICTIONS PER SAMPLE
#####################################################
if render_attention_cytometry_prediction_distributions_enabled:
out_dir_cytometry = out_dir + os.sep
# Oligos
try:
data_renderer.render_attention_cytometry_prediction_distributions(out_dir=out_dir_cytometry,
X_metadata=X_metadata,
title_suffix=' (Oligodendrocytes)',
y_preds=y_preds,
all_attentions=all_attentions,
filename_suffix='_oligo',
include_oligo=True,
include_neuron=False,
include_nucleus=False,
dpi=out_image_dpi)
data_renderer.render_attention_cytometry_prediction_distributions_partitioned(out_dir=out_dir_cytometry,
X_metadatas=X_metadata,
partitions=[2, 4],
title_suffix=' (Oligodendrocytes)',
y_preds=y_preds,
all_attentions=all_attentions,
filename_suffix='_oligo',
include_oligo=True,
include_neuron=False,
include_nucleus=False,
dpi=out_image_dpi)
except Exception as e:
log.write_exception(e)
# Neurons
try:
data_renderer.render_attention_cytometry_prediction_distributions(out_dir=out_dir_cytometry,
X_metadata=X_metadata,
title_suffix=' (Neurons)',
y_preds=y_preds,
all_attentions=all_attentions,
filename_suffix='_neurons',
include_oligo=False,
include_neuron=True,
include_nucleus=False,
dpi=out_image_dpi)
data_renderer.render_attention_cytometry_prediction_distributions_partitioned(out_dir=out_dir_cytometry,
X_metadatas=X_metadata,
partitions=[2, 4],
title_suffix=' (Neurons)',
y_preds=y_preds,
all_attentions=all_attentions,
filename_suffix='_neurons',
include_oligo=False,
include_neuron=True,
include_nucleus=False,
dpi=out_image_dpi)
except Exception as e:
log.write_exception(e)
# Nuclei
try:
data_renderer.render_attention_cytometry_prediction_distributions(out_dir=out_dir_cytometry,
X_metadata=X_metadata,
title_suffix=' (Nuclei)',
y_preds=y_preds,
all_attentions=all_attentions,
filename_suffix='_nuclei',
include_oligo=False,
include_neuron=False,
include_nucleus=True,
dpi=out_image_dpi)
data_renderer.render_attention_cytometry_prediction_distributions_partitioned(out_dir=out_dir_cytometry,
X_metadatas=X_metadata,
partitions=[2, 4],
title_suffix=' (Nuclei)',
y_preds=y_preds,
all_attentions=all_attentions,
filename_suffix='_nuclei',
include_oligo=False,
include_neuron=False,
include_nucleus=True,
dpi=out_image_dpi)
except Exception as e:
log.write_exception(e)
# All
try:
data_renderer.render_attention_cytometry_prediction_distributions(out_dir=out_dir_cytometry,
X_metadata=X_metadata,
title_suffix=' (All)',
y_preds=y_preds,
all_attentions=all_attentions,
filename_suffix='_all',
include_oligo=True,
include_neuron=True,
include_nucleus=True,
dpi=out_image_dpi)
data_renderer.render_attention_cytometry_prediction_distributions_partitioned(out_dir=out_dir_cytometry,
X_metadatas=X_metadata,
partitions=[2, 4],
title_suffix=' (All)',
y_preds=y_preds,
all_attentions=all_attentions,
filename_suffix='_all',
include_oligo=True,
include_neuron=True,
include_nucleus=True,
dpi=out_image_dpi)
except Exception as e:
log.write_exception(e)
#####################################################
# RENDERING ATTENTION CELL DISTRIBUTIONS
#####################################################
if render_attention_cell_distributions:
# calculating the cell count distribution for every attention value
distributions = mil_metrics.get_cells_per_attention(all_attentions, X_metadata)
data_renderer.render_attention_cell_distributions(out_dir=out_dir, distributions=distributions,
X_metadata=X_metadata, alpha=0.45,
title_suffix=' (Oligodendrocytes)', filename_suffix='_oligo',
include_oligo=True,
include_neuron=False,
include_nucleus=False,
dpi=out_image_dpi)
data_renderer.render_attention_cell_distributions(out_dir=out_dir, distributions=distributions,
X_metadata=X_metadata, alpha=0.45,
title_suffix=' (Neurons)', filename_suffix='_neuron',
include_oligo=False,
include_neuron=True,
include_nucleus=False,
dpi=out_image_dpi)
data_renderer.render_attention_cell_distributions(out_dir=out_dir, distributions=distributions,
X_metadata=X_metadata, alpha=0.45,
title_suffix=' (Nuclei)', filename_suffix='_nucleus',
include_oligo=False,
include_neuron=False,
include_nucleus=True,
dpi=out_image_dpi)
data_renderer.render_attention_cell_distributions(out_dir=out_dir, distributions=distributions,
X_metadata=X_metadata,
title_suffix=None, filename_suffix='_all',
include_oligo=True,
include_neuron=True,
include_nucleus=True,
dpi=out_image_dpi)
#####################################################
# RENDER ATTENTION HISTOGRAMS
#####################################################
data_renderer.render_attention_histograms(out_dir=out_dir, metadata_list=attention_metadata_list,
n_list=attention_n_list,
bins_list=attention_bins_list, otsu_index_list=attention_otsu_index_list,
otsu_threshold_list=attention_otsu_threshold_list,
entropy_attention_list=attention_entropy_attention_list,
entropy_hist_list=attention_entropy_hist_list, dpi=out_image_dpi * 2)
# Setting up result directories and file handles
experiment_names_unique = list(dict.fromkeys(experiment_names))
well_letters_unique_candidates = []
well_numbers_unique_candidates = []
handles = {}
for exp in experiment_names_unique:
handles[exp] = {}
# Removing previously existing paths, if they exist
if os.path.exists(out_dir + os.sep + exp + os.sep) and clear_old_data:
try:
shutil.rmtree(out_dir + os.sep + exp + os.sep)
except Exception as e:
log.write("Cannot remove path for " + exp + ". Msg: " + str(e))
del exp
# Setting up the instructions out file
sigmoid_instructions_file = out_dir + os.sep + 'sigmoid_instructions.csv'
if not os.path.exists(sigmoid_instructions_file):
f = open(sigmoid_instructions_file, 'w')
f.write('Experiment;Instructions: Dose;Instructions: Response')
f.close()
# Evaluating sigmoid performance using R
sigmoid_score_map = None
sigmoid_plot_estimation_map = None
sigmoid_plot_data_map = None
sigmoid_instructions_map = None
sigmoid_score_detail_map = None
sigmoid_bmc30_map = None
if r.has_connection(also_test_script=True) and render_dose_response_curves_enabled:
sigmoid_score_map, sigmoid_score_detail_map, sigmoid_plot_estimation_map, sigmoid_plot_data_map, sigmoid_instructions_map, sigmoid_bmc30_map = r.prediction_sigmoid_evaluation(
X_metadata=X_metadata, y_pred=y_hats, out_dir=out_dir, verbose=sigmoid_verbose,
save_sigmoid_plot=save_sigmoid_plot)
else:
log.write('Not running r evaluation. No connection.')
# Logging the instructions
if sigmoid_instructions_map is not None:
f = open(sigmoid_instructions_file, 'a')
for key in sigmoid_instructions_map.keys():
sigmoid_instructions = sigmoid_instructions_map[key]
f.write('\n' + key + ';' + sigmoid_instructions[0] + ';' + sigmoid_instructions[1])
del key, sigmoid_instructions
f.close()
# Rendering basic response curves
if render_dose_response_curves_enabled:
log.write('Rendering dose response curves.')
data_renderer.render_response_curves(X_metadata=X_metadata, y_pred=y_hats, out_dir=out_dir,
sigmoid_plot_estimation_map=sigmoid_plot_estimation_map,
sigmoid_plot_fit_map=sigmoid_plot_data_map,
sigmoid_score_detail_map=sigmoid_score_detail_map,
sigmoid_bmc30_map=sigmoid_bmc30_map,
sigmoid_score_map=sigmoid_score_map, dpi=int(out_image_dpi * 1.337))
# Rendering attention scores
if render_attention_spheres_enabled:
log.write('Rendering spheres and predictions.')
data_renderer.renderAttentionSpheres(X_raw=X_raw, X_metadata=X_metadata, input_dim=input_dim,
y_attentions=all_attentions, image_folder=image_folder, y_pred=y_hats,
render_merged_predicted_tiles_activation_overlays=render_merged_predicted_tiles_activation_overlays,
y_pred_binary=y_preds, out_dir=out_dir)
else:
log.write('Not rendering spheres and predictions.')
del X_raw
log.write('Finished rendering spheres.')
bars_width_mod = 10000
if sparse_hist:
bars_width_mod = 1000
print('\n') # new line so linux systems can write a single line
# Iterating over the predictions to save them to disc & dict
for i in range(len(y_hats)):
# Extracting predictions
# X_raw_current = X_raw[i]
X_metadata_current: [TileMetadata] = X_metadata[i]
experiment_names_current = experiment_names[i]
well_names_current = well_names[i]
all_attentions_current = all_attentions[i]
y_samples_pred_current = y_samples_pred[i]
y_hat_current = y_hats[i]
y_pred_current = y_preds[i]
line_print('[' + str(i + 1) + '/' + str(len(y_preds)) + '] Evaluating predictions for: ' +
experiment_names_current + ' - ' + well_names_current, include_in_log=True)
# Setting up handles & folders
current_exp_handle = handles[experiment_names_current]
if well_names_current not in current_exp_handle:
current_exp_handle[well_names_current] = {}
current_well_handle = current_exp_handle[well_names_current]
well_letter, well_number = extract_well_info(well_names_current, verbose=False)
well_letters_unique_candidates.append(well_letter)
well_numbers_unique_candidates.append(well_number)
# More Setting up handles & folders
all_attentions_current_normalized = all_attentions_current / max(all_attentions_current)
current_well_handle['attention'] = all_attentions_current
current_out_dir = out_dir + os.sep + experiment_names_current + os.sep + well_names_current + os.sep
os.makedirs(current_out_dir, exist_ok=True)
# Checking if metadata matches
assert X_metadata_current[0].experiment_name == experiment_names_current
assert X_metadata_current[0].well_letter.lower() in well_names_current.lower()
assert str(X_metadata_current[0].well_number) in well_names_current.lower()
# Choosing sparse / normalized data
all_attentions_used = all_attentions_current
if normalized_attention:
all_attentions_used = all_attentions_current_normalized
# Saving histogram
plt.clf()
# Saving raw data as CSV
f = open(current_out_dir + experiment_names_current + '-' + well_names_current + '-attention.csv', 'w')
f.write('Index;Attention;Attention (Normalized)\n')
for j in range(len(all_attentions_current)):
f.write(str(j) + ';')
f.write(str(all_attentions_current[j]) + ';' + str(all_attentions_used[j]))
f.write('\n')
f.write('Sum;' + str(sum(all_attentions_current)) + ';' + str(sum(all_attentions_used)))
f.close()
# writing the current handles back
current_exp_handle[well_names_current] = current_well_handle
handles[experiment_names_current] = current_exp_handle
del experiment_names_current, current_exp_handle
print('Done evaluating.')
well_letters_unique: [int] = list(dict.fromkeys(well_letters_unique_candidates))
well_numbers_unique: [str] = list(dict.fromkeys(well_numbers_unique_candidates))
well_letters_unique.sort()
well_numbers_unique.sort()
del well_letters_unique_candidates, well_numbers_unique_candidates
log.write('Finished predictions. Your results are here: ' + out_dir)
def generate_experiment_prediction_holders(X: [np.ndarray], experiment_names: [str], well_names: [str]):
# Reordering bags based on metadata
all_well_numbers = []
all_well_letters = []
experiment_holders = {}
for (bag, experiment_name, well_name) in zip(X, experiment_names, well_names):
if experiment_name not in experiment_holders.keys():
experiment_holders[experiment_name] = {}
current_holder = experiment_holders[experiment_name]
well_letter, well_number = loader.extract_well_info(well_name, verbose=False)
if well_number not in current_holder:
current_holder[well_number] = {}
if well_number not in all_well_numbers:
all_well_numbers.append(well_number)
if well_letter not in all_well_letters:
all_well_letters.append(well_letter)
current_holder[well_number][well_letter] = bag
experiment_holders[experiment_name] = current_holder
del current_holder
all_well_numbers.sort()
all_well_letters.sort()
# Iterating over all experiments and evaluating one at a time
experiment_names_unique = list(experiment_holders.keys())
experiment_names_unique.sort()
return experiment_holders, all_well_letters, all_well_numbers, experiment_names_unique
def main():
print('Predicting and creating a Dose Response curve for a whole bag.')
print('Platform:' + str(sys.platform))
debug = False
data_saver = True
render_attention_spheres_enabled = True
if sys.platform == 'win32':
debug = True
normalize_enum = None
model_path = default_out_dir_unix_base + os.sep + 'hnm-early_inverted-O3-adam-NoNeuron2-wells-normalize-7repack-0.65/'
if sys.platform == 'win32':
image_folder = paths.nucleus_predictions_image_folder_win
current_global_log_dir = 'U:\\bioinfdata\\work\\OmniSphero\\Sciebo\\HCA\\00_Logs\\mil_log\\win\\'
log.add_file('U:\\bioinfdata\\work\\OmniSphero\\Sciebo\\HCA\\00_Logs\\mil_log\\win\\all_logs.txt')
model_path = paths.windows_debug_model_path
normalize_enum = 6
else:
# good one:
model_path = '/mil/oligo-diff/models/linux/ep-aug-overlap-adadelta-endpoints-wells-normalize-6no-repack-round1/'
# testing:
model_path = '/mil/oligo-diff/models/linux/ep-aug-overlap-adadelta-endpoints-wells-normalize-8repack-0.5-round1/'
model_path = '/mil/oligo-diff/models/linux/ep-aug-overlap-adadelta-endpoints-wells-normalize-6repack-0.3-round1/'
# in production
model_path = '/mil/oligo-diff/models/production/paper_candidate_2/'
current_global_log_dir = '/Sciebo/HCA/00_Logs/mil_log3/linux-pred/'
image_folder = paths.nucleus_predictions_image_folder_unix
normalize_enum = 7
print('Log location: ' + current_global_log_dir)
print('Model path: ' + model_path)
assert os.path.exists(model_path)
checkpoint_file = model_path + os.sep + 'hnm' + os.sep + 'model_best.h5'
if not os.path.exists(checkpoint_file):
checkpoint_file = model_path + os.sep + 'model_best.h5'
assert os.path.exists(checkpoint_file)
if debug and sys.platform == 'win32':
predict_path(
model_save_path=model_path,
skip_already_predicted=False,
global_log_dir=current_global_log_dir,
checkpoint_file=checkpoint_file,
bag_paths=debug_prediction_dirs_win,
image_folder=image_folder,
data_loader_data_saver=True,
render_attention_spheres_enabled=render_attention_spheres_enabled,
channel_inclusions=loader.default_channel_inclusions_no_neurites,
render_merged_predicted_tiles_activation_overlays=False,
render_attention_histogram_enabled=False,
render_dose_response_curves_enabled=True,
predict_samples_as_bags=True,
used_tile_quartiles=None,
# oligo_positive_z_score_scale=2.0,
# oligo_z_score_max_kernel_size=10,
render_attention_instance_range_min=0.8,
render_attention_instance_range_max=1.0,
hist_bins_override=50,
sigmoid_verbose=True,
out_image_dpi=300,
render_attention_cytometry_prediction_distributions_enabled=True,
out_dir='U:\\bioinfdata\\work\\OmniSphero\\mil\\oligo-diff\\debug_predictions-win\\bmc\\',
gpu_enabled=False, normalize_enum=normalize_enum, max_workers=4)
elif sys.platform == 'win32':
checkpoint_file = 'U:\\bioinfdata\\work\\OmniSphero\\mil\\oligo-diff\\models\\linux\\ep-aug-overlap-adadelta-n-6-rp-0.3-l-mean_square_error-BMC\\model_best.h5'
for prediction_dir in paths.all_prediction_dirs_win:
predict_path(model_save_path=model_path,
global_log_dir=current_global_log_dir,
skip_already_predicted=False,
render_attention_spheres_enabled=render_attention_spheres_enabled,
render_attention_histogram_enabled=False,
render_merged_predicted_tiles_activation_overlays=False,
checkpoint_file=checkpoint_file,
out_dir=checkpoint_file + 'predictions-bmc\\',
bag_paths=[prediction_dir],
channel_inclusions=loader.default_channel_inclusions_no_neurites,
image_folder=image_folder,
hist_bins_override=50,
sigmoid_verbose=True,
used_tile_quartiles=None,
out_image_dpi=300,
render_attention_instance_range_min=0.8,
render_attention_instance_range_max=1.0,
render_dose_response_curves_enabled=True,
predict_samples_as_bags=False,
render_attention_cytometry_prediction_distributions_enabled=True,
gpu_enabled=False, normalize_enum=normalize_enum, max_workers=6)
else:
print('Predicting linux batches')
###########################################################
############## SETTING THE INPUT PATH HERE ################
###########################################################
# prediction_dirs_used = [paths.all_prediction_dirs_unix]
# prediction_dirs_used = [paths.curated_overlapping_source_dirs_unix_channel_transformed_rbg]
prediction_dirs_used = [paths.default_sigmoid_validation_dirs_unix]
###########################################################
###########################################################
# Setting the quartiles to be used
tile_quartiles = [
# [True, False, False, False]
# [False, True, False, False]
# [False, False, True, False]
# [False, False, False, True]
]
tile_quartiles = [loader.default_used_tile_quartiles]
###########################################################
if debug:
prediction_dirs_used = [prediction_dirs_used[0][0:3]]
if data_saver:
prediction_dirs_used = [[d] for d in prediction_dirs_used[0]]
prediction_dirs_used.sort()
time.sleep(2)
log.write('Number of plates to be predicted: ' + str(len(prediction_dirs_used)))
for i in range(len(prediction_dirs_used)):
prediction_dir = prediction_dirs_used[i]
log.write('#' + str(i + 1) + ': ' + str(prediction_dir))
del prediction_dir
time.sleep(0.69)
time.sleep(3)
for used_tile_quartiles in tile_quartiles:
used_tile_quartiles_enum = utils.boolean_to_integer(used_tile_quartiles[0],
used_tile_quartiles[1],
used_tile_quartiles[2],
used_tile_quartiles[3])
for i in range(len(prediction_dirs_used)):
prediction_dir = prediction_dirs_used[i]
log.write(
str(i + 1) + '/' + str(len(prediction_dirs_used)) + ' - Predicting: ' + str(prediction_dir))
log.clear_files()
if not type(prediction_dir) == list:
prediction_dir = [prediction_dir]
try:
out_dir_used = '/mil/oligo-diff/models/production/predictions/paper_candidate_2-quartile-AsBags'
os.makedirs(out_dir_used, exist_ok=True)
predict_path(checkpoint_file=checkpoint_file, model_save_path=model_path,
bag_paths=prediction_dir,
skip_already_predicted=False,
out_dir=out_dir_used,
global_log_dir=current_global_log_dir,
render_attention_spheres_enabled=render_attention_spheres_enabled,
render_merged_predicted_tiles_activation_overlays=False,
render_attention_histogram_enabled=True,
render_attention_cell_distributions=False,
render_dose_response_curves_enabled=True,
predict_samples_as_bags=True,
used_tile_quartiles=used_tile_quartiles,
render_attention_cytometry_prediction_distributions_enabled=False,
hist_bins_override=50,
out_image_dpi=800,
sigmoid_verbose=False,
render_attention_instance_range_min=0.9,
render_attention_instance_range_max=1.0,
image_folder=image_folder,
tile_constraints=loader.default_tile_constraints_none,
# tile_constraints=loader.default_tile_constraints_nuclei,
channel_inclusions=loader.default_channel_inclusions_no_neurites,
gpu_enabled=False, normalize_enum=normalize_enum, max_workers=20)
except Exception as e:
log.write('\n\n============================================================')
log.write('Fatal error during HT predictions: "' + str(e) + '"!')
log.write(str(e.__class__.__name__) + ': "' + str(e) + '"')
log.write_exception(e)
return
print('\n######################')
log.write('All Predictions done.')
log.write('Have a nice day. :)')
print('######################')
# Finishing up the logging process
log.write('Finished predicting & Dose-Response for all bags.')
if __name__ == '__main__':
main()