/
create_models.py
119 lines (105 loc) · 4.83 KB
/
create_models.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
import numpy as np
from sklearn.svm import OneClassSVM
from sklearn.decomposition import PCA
from sklearn.preprocessing import normalize
import argparse
import json
import pickle
import os
from quadrics import Quadrics
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("--methods",
nargs='+',
default=['OneClassSVM', 'PCA'],
help='Set empty to calculate OneClassSVM and PCA methods from config file')
args = parser.parse_args()
# we use embeddings normalization for training
print('Loading embeddings and config')
embeddings = normalize(np.load('image_embeddings/ms1m.npy'))
shuffle_indices = np.arange(len(embeddings))
np.random.shuffle(shuffle_indices)
with open("config.json", "r") as read_file:
config_file = json.load(read_file)
config_dict = config_file['train_params']
dir_models = config_file['models_dir']
if args.methods is None:
methods = config_dict.keys()
else: methods = args.methods
if not(os.path.isdir(dir_models)):
os.mkdir(dir_models)
if 'OneClassSVM' in methods:
print('OneClassSVM training...')
embs_indices = shuffle_indices[:config_dict['OneClassSVM']['n_points']]
config_dict['OneClassSVM'].pop('n_points')
clf = OneClassSVM(**config_dict['OneClassSVM']).fit(embeddings[embs_indices])
pickle.dump(clf, open(dir_models+'/OneClassSVM.pickle', 'wb'))
print('Model saved to '+dir_models+'/OneClassSVM.pickle')
if 'PCA' in methods:
print('PCA training...')
if config_dict['PCA']['n_points'] == 'all':
embs_indices = shuffle_indices
else:
embs_indices = shuffle_indices[:config_dict['PCA']['n_points']]
config_dict['PCA'].pop('n_points')
clf = PCA(**config_dict['PCA']).fit(embeddings[embs_indices])
pickle.dump(clf, open(dir_models+'/PCA.pickle', 'wb'))
print('Model saved to '+dir_models+'/PCA.pickle')
if 'quadrics' in methods:
print('Quadrics with type-2 distance training...')
if config_dict['quadrics']['n_points'] == 'all':
embs_indices = shuffle_indices
else:
embs_indices = shuffle_indices[:config_dict['quadrics']['n_points']]
n_quadrics = config_dict["quadrics"]["n_quadrics"]
distance = config_dict['quadrics']['distance']
lr = config_dict['quadrics']['lr']
n_epoch = config_dict['quadrics']['n_epoch']
device = config_dict['quadrics']['device']
batch_size = config_dict['quadrics']['batch_size']
val_size = config_dict['quadrics']['val_size']
clf = Quadrics(n_quadrics=n_quadrics, dist=distance, device=device)
if val_size > 0:
train_size = len(embeddings) - val_size
assert train_size > 0, "Validation size bigger than total length!!!"
val_dataset = embeddings[train_size:(train_size + val_size), :]
train_dataset = embeddings[:train_size, :]
else:
train_dataset = embeddings
val_dataset = None
clf.fit(train_dataset,
n_epoch,
learning_rate=lr,
batch_size=batch_size,
val_data=val_dataset)
clf.save(dir_models+'/Quadrics.pth')
print('Model saved to '+dir_models+'/Quadrics.pth')
if "quadrics_algebraic" in methods:
print('Quadrics with algebraic distance training...')
if config_dict['quadrics_algebraic']['n_points'] == 'all':
embs_indices = shuffle_indices
else:
embs_indices = shuffle_indices[:config_dict['quadrics_algebraic']['n_points']]
n_quadrics = config_dict["quadrics_algebraic"]["n_quadrics"]
distance = 'dist0'
lr = config_dict['quadrics_algebraic']['lr']
n_epoch = config_dict['quadrics_algebraic']['n_epoch']
device = config_dict['quadrics_algebraic']['device']
batch_size = config_dict['quadrics_algebraic']['batch_size']
val_size = config_dict['quadrics_algebraic']['val_size']
clf = Quadrics(n_quadrics=n_quadrics, dist=distance, device=device)
if val_size > 0:
train_size = len(embeddings) - val_size
assert train_size > 0, "Validation size bigger than total length!!!"
val_dataset = embeddings[train_size:(train_size + val_size), :]
train_dataset = embeddings[:train_size, :]
else:
train_dataset = embeddings
val_dataset = None
clf.fit(train_dataset,
n_epoch,
learning_rate=lr,
batch_size=batch_size,
val_data=val_dataset)
clf.save(dir_models+'/Quadrics_algebraic.pth')
print('Model saved to '+dir_models+'/Quadrics_algebraic.pth')