-
Notifications
You must be signed in to change notification settings - Fork 2
/
nontrain_selection.py
156 lines (117 loc) · 6.65 KB
/
nontrain_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
import torch
import torch.nn as nn
import numpy as np
import argparse
import random
import time
import open_clip
from open_clip import tokenizer, tokenize
from data import get_data, get_data_val
from text_preprocessing import text_preprocessing
def select_nontrain(args, target_model, preprocess_train, preprocess_val, device, length):
selected_nt_txt = []
selected_nt_url = []
selected_nt_cs_lst_tar = []
selected_nt_feat_lst_tar = []
args.val_data = args.val_data_nontrain_1
args.val_num_samples = args.val_num_samples_nontrain_1
valloader = get_data_val(args, (preprocess_val, preprocess_val))
valloader = valloader.dataloader
##################################################
### Nontrain data from valloader
start_time = time.time()
CC3M_LAION_commonset = np.load('./CC3M_LAION_commonset.npy')
CC3M_LAION_unqiue_commonset = np.load('./CC3M_LAION_unqiue_commonset.npy')
CC3M_LAION_url_commonset = np.load('./CC3M_LAION_url_commonset.npy')
cnt_nontrain = 0
for i, batch in enumerate( valloader ):
non_train_text = [text_preprocessing(q) for q in batch[1]]
non_train_url = [d['url'] for d in batch[2]]
common = np.intersect1d(np.array(non_train_text), CC3M_LAION_commonset)
x_ind = np.where(np.isin(np.array(non_train_text), common))[0]
common_2 = np.intersect1d(np.array(non_train_url), CC3M_LAION_url_commonset)
x_ind_2nd = np.where(np.isin(np.array(non_train_url), common_2))[0]
combined_x_ind = np.union1d(x_ind, x_ind_2nd)
sampled_ind = np.random.choice(np.arange(len(batch[1])), int(len(batch[1])*0.05), replace=False)
selected_ind = np.setdiff1d(sampled_ind, combined_x_ind)
images, texts = batch[0][selected_ind], tokenize(batch[1])[selected_ind]
cnt_nontrain += len(images)
images = images.to(device)
texts = texts.to(device)
with torch.no_grad(), torch.cuda.amp.autocast():
image_features2, text_features2, logit_scale2 = target_model(images, texts)
cs_2 = torch.diagonal( image_features2@text_features2.T )
selected_nt_url.extend( np.array(non_train_url)[selected_ind] )
selected_nt_txt.extend( np.array(non_train_text)[selected_ind] )
selected_nt_cs_lst_tar.extend( cs_2.detach().cpu().numpy() )
selected_nt_feat_lst_tar.extend( torch.cat([image_features2, text_features2], dim=1).detach().cpu() )
if cnt_nontrain >= length:
break
##################################################
## nontrain data | collect the data from the second val loader : CC12M
start_time = time.time()
CC12M_LAION_commonset = np.load('./CC12M_LAION_commonset.npy')
CC12M_LAION_unqiue_commonset = np.load('./CC12M_LAION_unqiue_commonset.npy')
CC12M_LAION_url_commonset = np.load('./CC12M_LAION_url_commonset.npy')
args.val_data = args.val_data_nontrain_2
args.val_num_samples = args.val_num_samples_nontrain_2
cc12m_valoader = get_data_val(args, (preprocess_val, preprocess_val))
cc12m_valoader = cc12m_valoader.dataloader
## nontrain data
cnt_nontrain = 0
for i, batch in enumerate( cc12m_valoader ):
non_train_text = [text_preprocessing(q) for q in batch[1]]
non_train_url = [d['url'] for d in batch[2]]
common = np.intersect1d(np.array(non_train_text), CC12M_LAION_commonset)
x_ind = np.where(np.isin(np.array(non_train_text), common))[0]
common_2 = np.intersect1d(np.array(non_train_url), CC12M_LAION_url_commonset)
x_ind_2nd = np.where(np.isin(np.array(non_train_url), common_2))[0]
combined_x_ind = np.union1d(x_ind, x_ind_2nd)
sampled_ind = np.random.choice(np.arange(len(batch[1])), int(len(batch[1])*0.05), replace=False)
selected_ind = np.setdiff1d(sampled_ind, combined_x_ind)
images, texts = batch[0][selected_ind], tokenize(batch[1])[selected_ind]
cnt_nontrain += len(images)
images = images.to(device)
texts = texts.to(device)
with torch.no_grad(), torch.cuda.amp.autocast():
image_features2, text_features2, logit_scale2 = target_model(images, texts)
cs_2 = torch.diagonal(image_features2@text_features2.T)
selected_nt_url.extend( np.array(non_train_url)[selected_ind] )
selected_nt_txt.extend( np.array(non_train_text)[selected_ind] )
selected_nt_cs_lst_tar.extend( cs_2.detach().cpu().numpy() )
selected_nt_feat_lst_tar.extend( torch.cat([image_features2, text_features2], dim=1).detach().cpu() )
if cnt_nontrain >= length:
break
##################################################
## nontrain data | collect the data from the second val loader : MSCOCO
args.val_data = args.val_data_nontrain_3
args.val_num_samples = args.val_num_samples_nontrain_3
mscoco_valoader = get_data_val(args, (preprocess_val, preprocess_val))
mscoco_valoader = mscoco_valoader.dataloader
start_time = time.time()
MSCOCO_LAION_commonset = np.load('./MSCOCO_LAION_commonset.npy')
MSCOCO_LAION_unqiue_commonset = np.load('./MSCOCO_LAION_unqiue_commonset.npy')
## nontrain data
cnt_nontrain = 0
for i, batch in enumerate( mscoco_valoader ):
non_train_text = [text_preprocessing(q) for q in batch[1]]
non_train_url = [d['url'] for d in batch[2]]
common = np.intersect1d(np.array(non_train_text), MSCOCO_LAION_commonset)
x_ind = np.where(np.isin(np.array(non_train_text), common))[0]
combined_x_ind = x_ind
sampled_ind = np.random.choice(np.arange(len(batch[1])), int(len(batch[1])*0.05), replace=False)
selected_ind = np.setdiff1d(sampled_ind, combined_x_ind)
images, texts = batch[0][selected_ind], tokenize(batch[1])[selected_ind]
cnt_nontrain += len(images)
images = images.to(device)
texts = texts.to(device)
with torch.no_grad(), torch.cuda.amp.autocast():
image_features2, text_features2, logit_scale2 = target_model(images, texts)
cs_2 = torch.diagonal( image_features2@text_features2.T )
selected_nt_url.extend( np.array(non_train_url)[selected_ind] )
selected_nt_txt.extend( np.array(non_train_text)[selected_ind] )
selected_nt_cs_lst_tar.extend( cs_2.detach().cpu().numpy() )
selected_nt_feat_lst_tar.extend( torch.cat([image_features2, text_features2], dim=1).detach().cpu() )
if cnt_nontrain >= length:
break
return selected_nt_txt, selected_nt_url, selected_nt_cs_lst_tar, selected_nt_feat_lst_tar