-
Notifications
You must be signed in to change notification settings - Fork 6
/
LOO_experiments_norb_clustering_sweep.py
120 lines (108 loc) · 4.38 KB
/
LOO_experiments_norb_clustering_sweep.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
"""
This performs clustering and then samples from those clusters in a stratified
manner
"""
import augmentations
import dataset_loaders
import experiments_util
import pretrained_experiments
import tensorflow as tf
from keras import backend as K
import pandas as pd
def main():
rounds = 5
n_aug_sample_points = [1, 10, 50, 100, 250, 500, 750, 1000]
n_train = 1000
n_jobs = 1
cv = 1
use_GPU = True
batch_size = 128
CNN_extractor_max_iter = 40
use_loss = False
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
K.set_session(sess)
# Can use multiple valus of C for cross-validation
logistic_reg__Cs = [[10]]
classes_datasets = [
((0, 1), dataset_loaders.Dataset.NORB),
]
selected_augmentations = [
(augmentations.Image_Transformation.translate, {"mag_aug": 6}),
(augmentations.Image_Transformation.rotate, {"mag_aug": 5,
"n_rotations": 4}),
(augmentations.Image_Transformation.crop, {"mag_augs": [2]}),
]
experiment_configs = [
("baseline", False, False),
("random_proportional", False, False),
("random_proportional", False, True),
("random_proportional", True, False),
("random_proportional", True, True),
("random_inverse_proportional", False, False),
# ("random_inverse_proportional", True, False),
# ("random_softmax_proportional", False, False),
# ("random_softmax_proportional", False, True),
# ("random_softmax_proportional", True, False),
# ("random_softmax_proportional", True, True),
# ("random_inverse_softmax_proportional", False, False),
# ("random_inverse_softmax_proportional", True, False),
("deterministic_proportional", False, False),
("deterministic_proportional", False, True),
("deterministic_proportional", True, False),
("deterministic_proportional", True, True),
("deterministic_inverse_proportional", False, False),
("deterministic_inverse_proportional", True, False),
]
for logistic_reg__C in logistic_reg__Cs:
for classes, dataset in classes_datasets:
for aug_transformation, aug_kw_args in selected_augmentations:
dataset_class_str = experiments_util.classes_to_class_str(
classes
)
print("Class types: {}".format(dataset_class_str))
reg_str = "-".join(list(map(str, logistic_reg__C)))
results_filename = "aug_results_{}_{}_{}_{}{}".format(
dataset.name,
dataset_class_str,
aug_transformation.name,
reg_str,
"_loss" if use_loss else "",
)
if dataset == dataset_loaders.Dataset.CIFAR10:
model_filename = "models/cifar10_ResNet56v2_model.h5"
elif dataset == dataset_loaders.Dataset.NORB:
if (aug_transformation ==
augmentations.Image_Transformation.translate):
model_filename = "models/norb_small_ResNet56v2_model_translate.h5"
else:
model_filename = \
"models/norb_small_ResNet56v2_model_rotate_crop.h5"
else:
raise RuntimeError("Unknown model for configuration")
all_results = pretrained_experiments.run_test_clustered_sweep(
classes,
rounds,
n_aug_sample_points,
n_train,
n_jobs,
cv,
use_GPU,
batch_size,
dataset,
aug_transformation,
aug_kw_args,
logistic_reg__C,
CNN_extractor_max_iter,
use_loss,
experiment_configs,
results_filename,
model_filename,
)
all_results_df = (pd.DataFrame(all_results)
.set_index("n_clusters"))
print("all_results", all_results_df)
all_results_df.to_csv(results_filename + "_sweep.csv")
if __name__ == "__main__":
main()