-
Notifications
You must be signed in to change notification settings - Fork 1
/
hyperparam_search.py
748 lines (659 loc) · 32.3 KB
/
hyperparam_search.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
import numpy as np
import os, sys
import matplotlib.pyplot as plt
from tqdm.autonotebook import tqdm, trange
import pandas as pd
import json
import itertools
import skopt
import skopt.plots
import argparse
from FewShotTestHandler import FewShotTestHandler, optimize_hyperparameters, find_hyperparameters, test_already_stored, filter_test_results
from dataset import DatasetHandler, get_valid_categories
from similarity_metrics import Similarity
from plotting_utils import plot
argparser = argparse.ArgumentParser()
argparser.add_argument("vlm", choices=["clip", "miles", "videoclip", "vifi"],
help="VLM to run. Requires corresponding conda environment")
argparser.add_argument("classifier", choices=["vl_proto", "hard_prompt_vl_proto", "nearest_neighbor", "gaussian_proto",
"linear", "subvideo", "tip_adapter", "coop", "cona", "cona_adapter",
"cona_prompt_init", "name_tuning", "name_tuning_adapter", "coop_adapter",
"vl_prompt", "vl_prompt_name_tuning"],
help="Classifier to run")
argparser.add_argument("-d", "--dataset", nargs="+", default=["smsm", "moma_sact", "kinetics_100", "moma_act", "hmdb_51", "ucf_101", "ssv2", "iadl"],
help="Which dataset name to run on")
argparser.add_argument("-s", "--n_shots", nargs="+", type=int, default=[1,2,4,8,16],
help="Number of shots to run on")
argparser.add_argument("-w", "--n_way", nargs="+", type=int, default=[None],
help="Number of categories to classify between. Default value None indicates choosing the max categories for each dataset.")
argparser.add_argument("--n_episodes", type=int, default=4,
help="Number of support set samples to repeat every test over")
argparser.add_argument("--val_tuning", type=lambda x: (x == "True"), default=True,
help="Whether or not the final trained classifier is reloaded from the epoch with the best val performance")
argparser.add_argument("--class-split", action="store_true",
help="Flag to use class-wise splitting (meta-learning paradigm) instead of video-wise splitting (finetuning paradigm)")
argparser.add_argument("-m", "--method", default="grid", choices=["gp", "forest", "random", "grid"],
help="Hyperparameter search method name.")
argparser.add_argument("-n", "--n_search_runs", type=int, default=32,
help="Sets the max number of hyperparameter search runs per test parameter value (dataset+n_shot)")
argparser.add_argument("-f", "--folder", default=None,
help="Optional folder path in which to save val and test results. By default creates folder for VLM and Classifier choice")
args, unknown_args_list = argparser.parse_known_args()
# Attempt to parse unknown args as vlm/classifier parameter overrides, like "--classifier.epochs 5 10 20"
cur_key = None
cur_key_vals = []
unknown_args = {}
for unknown_arg in unknown_args_list:
if unknown_arg.startswith("--"):
if cur_key is not None:
unknown_args[cur_key] = cur_key_vals
cur_key = unknown_arg[2:]
cur_key_vals = []
else:
if unknown_arg in ["True", "False"]:
unknown_arg = (unknown_arg == "True")
cur_key_vals.append(unknown_arg)
if cur_key is not None:
unknown_args[cur_key] = cur_key_vals
'''
Test Setup
'''
# Parameters which will be iterated over.
# Each product will receive individual hyperparam optimization over vlm and classifier params
test_params_dict = {}
# Dataset Params - dataset.____ keys are passed into DatasetHandler constructor
test_params_dict["dataset.name"] = args.dataset
# Few-Shot Test Params - test.____ keys are passed into few-shot test call
test_params_dict["test.n_way"] = args.n_way # None value gets manually converted to the max size for each dataset
test_params_dict["test.n_support"] = args.n_shots
test_params_dict["test.n_query"] = [None]
test_params_dict["test.n_episodes"] = [args.n_episodes]
'''
VLM Setup
'''
fixed_vlm_kwargs = {} # VLM keyword parameters to pass to constructor
vlm_hyperparams = [] # Hyperparameter spaces in skopt format
if args.vlm == "clip":
from CLIP.CLIPVLM import ClipVLM as VLM
fixed_vlm_kwargs["num_frames"] = 10
# Note: For better comparison with ViFi it may be worth using the vifi
# class for CLIP as well, ensuring the same CLIP backbone without the vifi checkpoint
#from VIFI_CLIP.wrapper import ViFiCLIP_SimilarityVLM as VLM
#fixed_vlm_kwargs["num_frames"] = 32
#fixed_vlm_kwargs["load_vifi"] = False
elif args.vlm == "miles":
from MILES.wrapper import MILES_SimilarityVLM as VLM
elif args.vlm == "videoclip":
from video_clip.video_clip import VideoClipVLM as VLM
fixed_vlm_kwargs["num_seconds"] = 4
fixed_vlm_kwargs["sample_strat"] = "spread"
fixed_vlm_kwargs["use_cuda"] = True
elif args.vlm == "univl":
from UNIVL.wrapper import UniVL_SimilarityVLM as VLM
elif args.vlm == "vttwins":
from VTTWINS.wrapper import VTTWINS_SimilarityVLM as VLM
elif args.vlm == "vifi":
from VIFI_CLIP.wrapper import ViFiCLIP_SimilarityVLM as VLM
fixed_vlm_kwargs["num_frames"] = 32
fixed_vlm_kwargs["load_vifi"] = True
else:
raise NotImplementedError
'''
Classifier Setup
'''
fixed_classifier_kwargs = {} # Classifier keyword parameters to pass to constructor
classifier_hyperparams = [] # Hyperparameter spaces in skopt format
if args.classifier == "vl_proto":
from classifier import WeightedTextFewShotClassifier as Classifier
classifier_hyperparams.append(skopt.space.Real(
1e-2, 1e2,
name="text_weight", prior="log-uniform"
))
classifier_hyperparams.append(skopt.space.Categorical(
["tip_adapter", "vid_action"],
name="prompt_ensemble_id"
))
elif args.classifier == "hard_prompt_vl_proto":
from classifier import HardPromptFewShotClassifier as Classifier
classifier_hyperparams.append(skopt.space.Real(
1e-2, 1e3,
name="text_weight", prior="log-uniform"
))
classifier_hyperparams.append(skopt.space.Categorical(
[
"a photo showing the activity of",
"a photo showing a",
"the video shows me",
"i am"
],
name="prompt_text"
))
elif args.classifier == "nearest_neighbor":
from classifier import NearestNeighborFewShotClassifier as Classifier
classifier_hyperparams.append(skopt.space.Integer(
1, 32,
name="neighbor_count", prior="log-uniform"
))
classifier_hyperparams.append(skopt.space.Categorical(
["uniform", "distance"],
name="neighbor_weights"
))
elif args.classifier == "gaussian_proto":
from classifier import GaussianFewShotClassifier as Classifier
classifier_hyperparams.append(skopt.space.Real(
1e-2, 1000,
name="text_weight", prior="log-uniform"
))
classifier_hyperparams.append(skopt.space.Integer(
0, 100,
name="prior_count", prior="uniform"
))
classifier_hyperparams.append(skopt.space.Real(
1e-2, 1e2,
name="prior_var", prior="log-uniform"
))
elif args.classifier == "linear":
from classifier import LinearProbe as Classifier
classifier_hyperparams.append(skopt.space.Real(
1e-2, 1e2,
name="regularization", prior="log-uniform"
))
elif args.classifier == "subvideo":
from classifier import SubVideoAverageFewShotClassifier as Classifier
elif args.classifier == "tip_adapter":
from classifier import TipAdapterFewShotClassifier as Classifier
fixed_classifier_kwargs["finetune_epochs"] = 20
fixed_classifier_kwargs["random_augment"] = False
fixed_classifier_kwargs["beta"] = 5.5
classifier_hyperparams.append(skopt.space.Categorical(
[1e0, 1e1, 1e2, 1e3],
name="alpha"
))
classifier_hyperparams.append(skopt.space.Categorical(
[1e-4, 4e-4, 1e-3, 4e-3],
name="finetune_lr"
))
classifier_hyperparams.append(skopt.space.Categorical(
["tip_adapter", "vid_action"],
name="prompt_ensemble_id"
))
elif args.classifier == "coop":
from classifier import CoopFewShotClassifier as Classifier
fixed_classifier_kwargs["random_augment"] = False
fixed_classifier_kwargs["batch_size"] = 8
fixed_classifier_kwargs["optimizer"] = "sgd"
fixed_classifier_kwargs["epochs"] = 50
fixed_classifier_kwargs["csc"] = False
classifier_hyperparams.append(skopt.space.Categorical(
[8, 16],
name="context_len"
))
classifier_hyperparams.append(skopt.space.Categorical(
[6.25e-5, 5e-4, 2e-3, 4e-3],
name="lr"
))
elif args.classifier == "cona":
from classifier import CoNaFewShotClassifier as Classifier
fixed_classifier_kwargs["random_augment"] = False
fixed_classifier_kwargs["batch_size"] = 8
fixed_classifier_kwargs["optimizer"] = "adamw"
fixed_classifier_kwargs["epochs"] = 20
fixed_classifier_kwargs["context_len"] = 4
classifier_hyperparams.append(skopt.space.Categorical(
[1e-5],
name="lr"
))
classifier_hyperparams.append(skopt.space.Categorical(
[1, 5, 10, 20],
name="name_regularization"
))
elif args.classifier == "cona_prompt_init":
from classifier import CoNaPromptInitFewShotClassifier as Classifier
fixed_classifier_kwargs["random_augment"] = False
fixed_classifier_kwargs["batch_size"] = 8
fixed_classifier_kwargs["optimizer"] = "adamw"
fixed_classifier_kwargs["epochs"] = 20
fixed_classifier_kwargs["context_prompt_init"] = "a photo of"
classifier_hyperparams.append(skopt.space.Categorical(
[1e-5],
name="lr"
))
classifier_hyperparams.append(skopt.space.Categorical(
[10],
name="name_regularization"
))
classifier_hyperparams.append(skopt.space.Categorical(
[10],
name="context_regularization"
))
elif args.classifier == "cona_adapter":
from classifier import CoNaAdapterFewShotClassifier as Classifier
fixed_classifier_kwargs["random_augment"] = False
fixed_classifier_kwargs["batch_size"] = 8
fixed_classifier_kwargs["optimizer"] = "adamw"
fixed_classifier_kwargs["epochs"] = 20
#fixed_classifier_kwargs["name_regularization"] = 20
fixed_classifier_kwargs["adapter_regularization"] = 1e-2
classifier_hyperparams.append(skopt.space.Categorical(
[1e-3],
name="lr"
))
classifier_hyperparams.append(skopt.space.Categorical(
[0.1],
name="adapter_lr_multiplier"
))
classifier_hyperparams.append(skopt.space.Categorical(
[1, 2, 4],
name="context_len"
))
classifier_hyperparams.append(skopt.space.Categorical(
[20],
name="name_regularization"
))
elif args.classifier == "name_tuning":
from classifier import NameTuningFewShotClassifier as Classifier
fixed_classifier_kwargs["random_augment"] = False
fixed_classifier_kwargs["batch_size"] = 8
fixed_classifier_kwargs["optimizer"] = "adamw"
fixed_classifier_kwargs["epochs"] = 20
classifier_hyperparams.append(skopt.space.Categorical(
[1e-3, 4e-3, 1e-4, 1e-5],
name="lr"
))
classifier_hyperparams.append(skopt.space.Categorical(
[0.01, 0.1, 1, 10],
name="name_regularization"
))
classifier_hyperparams.append(skopt.space.Categorical(
["simple_video"],
name="prompt_ensemble_id"
))
elif args.classifier == "name_tuning_adapter":
from classifier.name_tuning_adapter import NameTuningAdapterFewShotClassifier as Classifier
fixed_classifier_kwargs["random_augment"] = False
fixed_classifier_kwargs["batch_size"] = 8
fixed_classifier_kwargs["optimizer"] = "adamw"
fixed_classifier_kwargs["epochs"] = 20
fixed_classifier_kwargs["low_memory_training"] = True
fixed_classifier_kwargs["lr"] = 1e-4
fixed_classifier_kwargs["adapter_lr_multiplier"] = 1e-1
fixed_classifier_kwargs["name_regularization"] = 20
fixed_classifier_kwargs["adapter_regularization"] = 1e-2
fixed_classifier_kwargs["prompt_ensemble_id"] = "tip_adapter"
elif args.classifier == "coop_adapter":
from classifier.coop_adapter import CoopAdapterFewShotClassifier as Classifier
fixed_classifier_kwargs["random_augment"] = False
fixed_classifier_kwargs["batch_size"] = 8
fixed_classifier_kwargs["optimizer"] = "sgd"
fixed_classifier_kwargs["epochs"] = 20
fixed_classifier_kwargs["lr"] = 2e-3
fixed_classifier_kwargs["adapter_lr_multiplier"] = 1e-1
fixed_classifier_kwargs["adapter_regularization"] = 1e-2
elif args.classifier == "vl_prompt":
from classifier.vl_prompt import VLPromptFewShotClassifier as Classifier
fixed_classifier_kwargs["batch_size"] = 2
fixed_classifier_kwargs["accumulation_steps"] = 4
elif args.classifier == "vl_prompt_name_tuning":
from classifier.vl_prompt_name_tuning import VLPromptNameTuningFewShotClassifier as Classifier
fixed_classifier_kwargs["batch_size"] = 2
fixed_classifier_kwargs["accumulation_steps"] = 4
fixed_classifier_kwargs["name_regularization"] = 100
classifier_hyperparams.append(skopt.space.Categorical(
["adamw", "sgd"],
name="optimizer"
))
classifier_hyperparams.append(skopt.space.Categorical(
[0.1, 1.0],
name="name_lr_multiplier"
))
else:
raise ValueError("Unrecognized classifier arg")
# Update classifier/vlm args with any that have been manually specified
# If only a single value follows, fixed_<classifier/vlm>_args is updated,
# if multiple values follow, <classifier/vlm>_hyperparams are updated or added to
for key, val_list in unknown_args.items():
if key.startswith("vlm.") and len(val_list) > 0:
key = key[4:]
if len(val_list) == 1:
# Update fixed args
fixed_vlm_kwargs[key] = val_list[0]
else:
# Update hyperparams
new_hyperparam = skopt.space.Categorical(val_list, name=key)
matched_name_hyperparam_inds = [i for i, hyper in enumerate(vlm_hyperparams) if hyper.name == key]
if len(matched_name_hyperparam_inds):
vlm_hyperparams[matched_name_hyperparam_inds[0]] = new_hyperparam
else:
vlm_hyperparams.append(new_hyperparam)
elif key.startswith("classifier.") and len(val_list) > 0:
key = key[11:]
if len(val_list) == 1:
# Update fixed args
fixed_classifier_kwargs[key] = val_list[0]
else:
# Update hyperparams
new_hyperparam = skopt.space.Categorical(val_list, name=key)
matched_name_hyperparam_inds = [i for i, hyper in enumerate(classifier_hyperparams) if hyper.name == key]
if len(matched_name_hyperparam_inds):
classifier_hyperparams[matched_name_hyperparam_inds[0]] = new_hyperparam
else:
classifier_hyperparams.append(new_hyperparam)
elif key.startswith("test.") or key.startswith("dataset."):
test_params_dict[key] = val_list
else:
raise ValueError(f"Unrecognized argument: --{' '.join([key] + val_list)}")
'''
Set up results folder
'''
if args.folder is None:
RESULTS_FOLDER = os.path.join("hyperparam_search", args.classifier, args.vlm)
else:
RESULTS_FOLDER = args.folder
os.makedirs(RESULTS_FOLDER, exist_ok=True)
val_run_handler = FewShotTestHandler(os.path.join(RESULTS_FOLDER, "results_val.csv"))
test_run_handler = FewShotTestHandler(os.path.join(RESULTS_FOLDER, "results_test.csv"))
'''
Hyperparameter Search
'''
# Combine vlm and classifier hyperparams
for vlm_hyper in vlm_hyperparams:
vlm_hyper.name = f"vlm.{vlm_hyper.name}"
for classifier_hyper in classifier_hyperparams:
classifier_hyper.name = f"classifier.{classifier_hyper.name}"
hyperparam_space = vlm_hyperparams + classifier_hyperparams
support_dataset_val = None # Source for example videos during val phase
query_dataset_val = None # Source for query videos during val phase
support_dataset_test = None # Source for example videos during test phase
query_dataset_test = None # Source for query videos during test phase
val_tuning_dataset = None # Optional set of val query videos for selecting best epoch (in val and test phase)
cur_dataset_kwargs = None
vlm = None
cur_vlm_kwargs = None
pbar = tqdm(list(itertools.product(*test_params_dict.values())))
for test_params in pbar:
test_params = dict(zip(test_params_dict.keys(), test_params))
pbar.set_postfix(test_params)
dataset_kwargs = {key[8:]: val for key, val in test_params.items() if key.startswith("dataset.")}
test_kwargs = {key[5:]: val for key, val in test_params.items() if key.startswith("test.")}
# Update datasets
if cur_dataset_kwargs != dataset_kwargs:
# Primary case: fine-tuning paradigm, videowise splits, support set = train split, query set = val or test split
if not args.class_split:
query_dataset_val = DatasetHandler(**dataset_kwargs, split="val", split_type="video")
query_dataset_test = DatasetHandler(**dataset_kwargs, split="test", split_type="video")
support_dataset_val = support_dataset_test = DatasetHandler(**dataset_kwargs, split="train", split_type="video")
val_tuning_dataset = query_dataset_val if args.val_tuning else None
# Alternate case: meta-learning paradigm, classwise splits, support set + query set drawn from same split
# val-tuning is disabled with this setting
else:
query_dataset_val = support_dataset_val = DatasetHandler(**dataset_kwargs, split="val", split_type="class")
query_dataset_test = support_dataset_test = DatasetHandler(**dataset_kwargs, split="test", split_type="class")
val_tuning_dataset = None
cur_dataset_kwargs = dataset_kwargs
# Convert n_way = None into n_way = max-ways (ensuring each category has enough datapoints per split)
if test_kwargs["n_way"] is None:
test_kwargs["n_way"] = min(
len(get_valid_categories(query_dataset_val, support_dataset_val, test_kwargs["n_support"], test_kwargs["n_query"], val_tuning_dataset)),
len(get_valid_categories(query_dataset_test, support_dataset_test, test_kwargs["n_support"], test_kwargs["n_query"], val_tuning_dataset))
)
'''
Define functions for skopt optimizer methods, or for generally running tests with any sampled hyperparam values
'''
# skopt loss function
@skopt.utils.use_named_args(hyperparam_space)
def val_neg_accuracy(**hyperparam_kwargs):
hyperparam_kwargs = dict(hyperparam_kwargs)
vlm_kwargs = {key[4:]: val for key, val in hyperparam_kwargs.items() if key.startswith("vlm.")}
classifier_kwargs = {key[11:]: val for key, val in hyperparam_kwargs.items() if key.startswith("classifier.")}
# Update vlm if necessary (allow reuse if unchanging)
global vlm, cur_vlm_kwargs
if vlm is None or cur_vlm_kwargs != vlm_kwargs:
vlm = VLM(**dict(fixed_vlm_kwargs, **vlm_kwargs))
cur_vlm_kwargs = vlm_kwargs
# Update classifier
classifier = Classifier(vlm, **dict(fixed_classifier_kwargs, **classifier_kwargs))
accuracy = val_run_handler.run_few_shot_test(classifier, query_dataset_val, support_dataset_val, **test_kwargs, val_tuning_dataset=val_tuning_dataset)
return -1 * accuracy
# Callback function for progress bar
hyper_search_pbar = None
def skopt_callback(current_skopt_results):
best_run_ind = np.argmin(current_skopt_results.func_vals)
postfix_dict = {
"best_acc": round(-1 * current_skopt_results.func_vals[best_run_ind], 5)
}
for i, param_space in enumerate(hyperparam_space):
key = param_space.name
val = current_skopt_results.x_iters[best_run_ind][i]
if isinstance(val, float):
val = round(val, 5)
postfix_dict[key] = val
hyper_search_pbar.set_postfix(postfix_dict)
hyper_search_pbar.update(1)
'''
Potentially Skip if matching run already exists in the test split results csv
'''
matching_test_run_results = filter_test_results(
test_run_handler.results,
dict(
test_kwargs,
query_dataset=query_dataset_test.id(),
support_dataset=support_dataset_test.id(),
val_tuning_dataset=val_tuning_dataset.id() if val_tuning_dataset is not None else None,
vlm_class=VLM.__name__,
**{f"vlm.{key}": val for key, val in fixed_vlm_kwargs.items()},
classifier_class=Classifier.__name__,
**{f"classifier.{key}": val for key, val in fixed_classifier_kwargs.items()}
)
)
if len(matching_test_run_results):
print(f"Skipping hyperparam search which already has test results.")
print(f"Dataset: {query_dataset_test.id()}")
print(f"Test kwargs:\n{json.dumps(test_kwargs, indent=2)}")
continue
'''
Run hyperparameter search for current test params.
SKIP if there are no hyperparameter spaces defined for the current setup
'''
if len(hyperparam_space) > 0:
if args.method in ["gp", "forest"]:
'''
Find any previous val runs which shall be fed into skopt hyperparam search alg
Possible since hyperparameter spaces are named to match names in results csvs, which cover all vlm and classifier parameters
Only used for skopt search methods
'''
prev_val_run_results = filter_test_results(
val_run_handler.results,
dict(
test_kwargs,
query_dataset=query_dataset_val.id(),
support_dataset=support_dataset_val.id(),
val_tuning_dataset=val_tuning_dataset.id() if val_tuning_dataset is not None else None,
vlm_class=VLM.__name__,
**{f"vlm.{key}": val for key, val in fixed_vlm_kwargs.items()},
classifier_class=Classifier.__name__,
**{f"classifier.{key}": val for key, val in fixed_classifier_kwargs.items()}
)
).reset_index(drop=True)
if len(prev_val_run_results):
x0, y0 = [], []
for i in range(len(prev_val_run_results)):
# Discard points outside of the current hyperparam space bounds
in_bounds = True
for hyper in hyperparam_space:
val = prev_val_run_results.loc[i, hyper.name]
if isinstance(hyper, skopt.space.space.Categorical):
if val not in hyper.categories:
in_bounds = False
break
else:
if val < hyper.low or val > hyper.high:
in_bounds = False
break
if in_bounds:
x0.append(tuple(prev_val_run_results.loc[i, hyper.name] for hyper in hyperparam_space))
y0.append(-1 * prev_val_run_results.loc[i, "accuracy"])
else:
x0, y0 = None, None
if args.method == "gp":
skopt_func = skopt.gp_minimize
elif args.method == "forest":
skopt_func = skopt.forest_minimize
else:
raise NotImplementedError
hyper_search_pbar = tqdm(total=args.n_search_runs)
skopt_results = skopt_func(val_neg_accuracy, hyperparam_space, n_calls=args.n_search_runs, callback=skopt_callback, x0=x0, y0=y0)
elif args.method == "random":
hyper_search_pbar = tqdm(total=args.n_search_runs)
for _ in range(args.n_search_runs):
val_neg_accuracy([hyper.rvs(1)[0] for hyper in hyperparam_space])
hyper_search_pbar.update(1)
elif args.method == "grid":
'''
If doing grid search, convert categorical and continues hyperparam spaces into an evenly spaced grid of samples
'''
categorical_hyperparams = [hyper for hyper in hyperparam_space if type(hyper) is skopt.space.space.Categorical]
other_hyperparams = [hyper for hyper in hyperparam_space if type(hyper) is not skopt.space.space.Categorical]
# Grid must iterate over all selected categories
runs_per_category_choice = args.n_search_runs
for hyper in categorical_hyperparams:
runs_per_category_choice = runs_per_category_choice // len(hyper.categories)
if runs_per_category_choice == 0:
raise ValueError(f"Too many categorical hyperparameters to iterate over all choices without exceeding {args.n_search_runs} runs.")
if len(other_hyperparams) == 0:
discretized_hyperparam_space = [hyper.categories for hyper in hyperparam_space]
else:
samples_per_cont_hyper = int(np.power(runs_per_category_choice, 1 / len(other_hyperparams)))
if samples_per_cont_hyper == 0:
raise ValueError(f"Too many hyperparameters to iterate over all categories and still choose multiple values per continuous space, without exceeding {args.n_search_runs} runs.")
discretized_hyperparam_space = []
for hyper in hyperparam_space:
if type(hyper) is skopt.space.space.Categorical:
discretized_hyperparam_space.append(hyper.categories)
elif type(hyper) in [skopt.space.space.Real, skopt.space.space.Integer]:
if hyper.prior == "log-uniform":
hyper_samples = np.logspace(np.log10(hyper.low), np.log10(hyper.high), num=samples_per_cont_hyper, endpoint=True)
else:
hyper_samples = np.linspace(hyper.low, hyper.high, num=samples_per_cont_hyper, endpoint=True)
if type(hyper) is skopt.space.space.Integer:
hyper_samples = np.round(hyper_samples)
discretized_hyperparam_space.append(hyper_samples)
else:
raise NotImplementedError
hyperparam_value_iter = list(itertools.product(*discretized_hyperparam_space))
hyper_search_pbar = tqdm(total=len(hyperparam_value_iter))
for i, hyperparam_values in enumerate(hyperparam_value_iter):
val_neg_accuracy(hyperparam_values)
hyper_search_pbar.update(1)
else:
raise NotImplementedError
'''
Save skopt dependence plots if not grid search
'''
if args.method != "grid":
completed_val_run_results = filter_test_results(
val_run_handler.results,
dict(
test_kwargs,
query_dataset=query_dataset_val.id(),
support_dataset=support_dataset_val.id(),
val_tuning_dataset=val_tuning_dataset.id() if val_tuning_dataset is not None else None,
vlm_class=VLM.__name__,
**{f"vlm.{key}": val for key, val in fixed_vlm_kwargs.items()},
classifier_class=Classifier.__name__,
**{f"classifier.{key}": val for key, val in fixed_classifier_kwargs.items()}
)
).reset_index(drop=True)
x0, y0 = [], []
for i in range(len(completed_val_run_results)):
# Discard points outside of the current hyperparam space bounds
in_bounds = True
for hyper in hyperparam_space:
val = completed_val_run_results.loc[i, hyper.name]
if isinstance(hyper, skopt.space.space.Categorical):
if val not in hyper.categories:
in_bounds = False
break
else:
if val < hyper.low or val > hyper.high:
in_bounds = False
break
if in_bounds:
x0.append(tuple(completed_val_run_results.loc[i, hyper.name] for hyper in hyperparam_space))
y0.append(-1 * completed_val_run_results.loc[i, "accuracy"])
dummy_results = skopt.gp_minimize(val_neg_accuracy, hyperparam_space, n_calls=0, n_initial_points=0, x0=x0, y0=y0)
ax = skopt.plots.plot_objective(dummy_results, levels=100)
if isinstance(ax, np.ndarray):
fig = ax[0,0].get_figure()
else:
fig = ax.get_figure()
fig.savefig(
os.path.join(RESULTS_FOLDER, ".".join([query_dataset_val.id()] + [f"{key}_{val}" for key, val in test_kwargs.items()] + ["jpg"])),
facecolor="white", bbox_inches="tight"
)
'''
Collect best hyperparam values from val results
'''
vlm_kwargs = {}
classifier_kwargs = {}
if len(hyperparam_space) > 0:
# Select best hyperparameter values from val split
best_hyperparam_values = find_hyperparameters(
val_run_handler.results,
hyperparam_cols=[col for col in val_run_handler.results if col.startswith("classifier.") or col.startswith("vlm.")]
)
matching_hyperparam_values = filter_test_results(
best_hyperparam_values,
dict(
test_kwargs,
query_dataset=query_dataset_val.id(),
support_dataset=support_dataset_val.id(),
val_tuning_dataset=val_tuning_dataset.id() if val_tuning_dataset is not None else None,
vlm_class=VLM.__name__,
**{f"vlm.{key}": val for key, val in fixed_vlm_kwargs.items()},
classifier_class=Classifier.__name__,
**{f"classifier.{key}": val for key, val in fixed_classifier_kwargs.items()}
)
).reset_index(drop=True)
for col in matching_hyperparam_values.columns:
# Skip vlm/classifier args that aren't in hyperparam space
if col not in [hyper.name for hyper in hyperparam_space]:
continue
if col.startswith("vlm."):
if col[4:] in fixed_vlm_kwargs.keys():
continue
val = matching_hyperparam_values.loc[0, col]
# NaN values indicate they aren't applicable for this vlm/classifier
if not pd.isna(val):
# Replace np types with native python types
if type(val).__module__ == np.__name__:
val = val.item()
vlm_kwargs[col[4:]] = val
if col.startswith("classifier."):
if col[11:] in fixed_classifier_kwargs.keys():
continue
val = matching_hyperparam_values.loc[0, col]
# NaN values indicate they aren't applicable for this vlm/classifier
if not pd.isna(val):
# Replace np types with native python types
if type(val).__module__ == np.__name__:
val = val.item()
if col == "classifier.metric":
classifier_kwargs[col[11:]] = Similarity[val]
else:
classifier_kwargs[col[11:]] = val
'''
Run on test set with best hyperparameters
'''
# Update vlm if necessary (allow reuse if unchanging)
if vlm is None or cur_vlm_kwargs != vlm_kwargs:
vlm = VLM(**fixed_vlm_kwargs, **vlm_kwargs)
cur_vlm_kwargs = vlm_kwargs
# Update classifier
classifier = Classifier(vlm, **fixed_classifier_kwargs, **classifier_kwargs)
test_acc = test_run_handler.run_few_shot_test(classifier, query_dataset_test, support_dataset_test, **test_kwargs, val_tuning_dataset=val_tuning_dataset)
print(f"Test Run Complete!")
print(f"Accuracy: {test_acc}")
print(f"Dataset: {query_dataset_test.id()}")
print(f"Test: {json.dumps(test_kwargs, indent=2)}")
print(f"VLM: {json.dumps(vlm.params(), indent=2)}")
print(f"Classifier: {json.dumps(classifier.params(), indent=2)}")