forked from billpsomas/metrix
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
165 lines (116 loc) · 6.78 KB
/
train.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
import torch
from pytorch_metric_learning.utils import loss_and_miner_utils as lmu
import random
from general_utils import *
from mixup.utils import *
from mixup.losses import *
def baseline_contrastive(inputs, target, model, distance, reducer, opt, losses_per_epoch):
out = model(inputs)
embedding_similarity = distance(out)
a1, p, a2, n = lmu.get_all_pairs_indices(target)
pos_pair_dist, neg_pair_dist = [], []
if len(a1) > 0:
pos_pair_dist = embedding_similarity[a1, p]
if len(a2) > 0:
neg_pair_dist = embedding_similarity[a2, n]
indices_tuple = (a1, p, a2, n)
pos_pairs = lmu.pos_pairs_from_tuple(indices_tuple)
neg_pairs = lmu.neg_pairs_from_tuple(indices_tuple)
if len(pos_pair_dist) > 0:
pos_loss = torch.nn.functional.relu(margin(pos_pair_dist, 0.0))
if len(neg_pair_dist) > 0:
neg_loss = torch.nn.functional.relu(margin(0.5, neg_pair_dist))
loss_dict = {
"pos_loss": {
"losses": pos_loss,
"indices": pos_pairs,
"reduction_type": "pos_pair",
},
"neg_loss": {
"losses": neg_loss,
"indices": neg_pairs,
"reduction_type": "neg_pair",
},
}
loss = reducer(loss_dict, embedding_similarity, target)
opt.zero_grad()
loss.backward()
torch.nn.utils.clip_grad_value_(model.parameters(), 10)
losses_per_epoch.append(loss.data.cpu().numpy())
opt.step()
return loss, losses_per_epoch
def input_metrix_contrastive(inputs, target, model, distance, criterion, opt, losses_per_epoch, reducer_pos, reducer_neg):
opt.zero_grad()
out_clean = model(inputs)
loss_clean = criterion(out_clean, target)
layer_mix = random.randint(0,1)
a1, pos, a2, neg = lmu.get_all_pairs_indices(target)
if (layer_mix == 0):
sim_mix, y_pos, y_neg, new_a2, pos, new_neg, lam_pn = input_posneg_pair_mixup_for_pos_anchor(a1, pos, a2, neg, out_clean, inputs, target, distance, model, top_k=1)
loss_pos_dict, loss_neg_dict = loss_posneg_mixup(sim_mix, pos, new_a2, new_neg, lam_pn)
loss_mixed_posneg_for_pos_anc = reducer_pos(loss_pos_dict, sim_mix, y_pos) + reducer_neg(loss_neg_dict, sim_mix, y_neg)
loss = loss_clean + 0.4*loss_mixed_posneg_for_pos_anc
elif (layer_mix == 1):
sim_mix_ancneg, y_pos_an, y_neg_an, new_a2_an, new_neg_an, lam_an = input_neg_pair_mixup_without_posanchor(a1, a2, neg, out_clean, inputs, target, distance, model, top_k=1)
loss_dict_pos_ancneg, loss_dict_neg_ancneg = loss_posneg_mixup(sim_mix_ancneg, new_a2_an, new_a2_an, new_neg_an, lam_an)
loss_mixed_ancneg_without_pos_anc = reducer_pos(loss_dict_pos_ancneg, sim_mix_ancneg, y_pos_an) + reducer_neg(loss_dict_neg_ancneg, sim_mix_ancneg, y_neg_an)
loss = loss_clean + 0.4*loss_mixed_ancneg_without_pos_anc
loss.backward()
torch.nn.utils.clip_grad_value_(model.parameters(), 10)
losses_per_epoch.append(loss.data.cpu().numpy())
opt.step()
return loss, losses_per_epoch
def embed_metrix_contrastive(inputs, target, model, distance, criterion, opt, losses_per_epoch, reducer_pos, reducer_neg):
opt.zero_grad()
out_clean = model(inputs)
loss_clean = criterion(out_clean, target)
layer_mix = random.randint(0,1)
a1, pos, a2, neg = lmu.get_all_pairs_indices(target)
if (layer_mix == 0):
sim_mix, y_pos, y_neg, new_a2, new_pos, new_neg, lam_pn = embed_posneg_pair_mixup_for_pos_anchor(a1, pos, a2, neg, out_clean, inputs, target, distance)
loss_pos_dict, loss_neg_dict = loss_posneg_mixup(sim_mix, new_pos, new_a2, new_neg, lam_pn)
loss_mixed_posneg_for_pos_anc = reducer_pos(loss_pos_dict, sim_mix, y_pos) + reducer_neg(loss_neg_dict, sim_mix, y_neg)
loss = loss_clean + 0.4*loss_mixed_posneg_for_pos_anc
elif (layer_mix == 1):
sim_mix_ancneg, y_pos_an, y_neg_an, new_a2_an, new_neg_an, lam_an = embed_neg_pair_mixup_without_posanchor(a1, a2, neg, out_clean, inputs, target, distance)
loss_dict_pos_ancneg, loss_dict_neg_ancneg = loss_posneg_mixup(sim_mix_ancneg, new_a2_an, new_a2_an, new_neg_an, lam_an)
loss_mixed_ancneg_without_pos_anc = reducer_pos(loss_dict_pos_ancneg, sim_mix_ancneg, y_pos_an) + reducer_neg(loss_dict_neg_ancneg, sim_mix_ancneg, y_neg_an)
loss = loss_clean + 0.3*loss_mixed_ancneg_without_pos_anc
loss.backward()
torch.nn.utils.clip_grad_value_(model.parameters(), 10)
losses_per_epoch.append(loss.data.cpu().numpy())
opt.step()
return loss, losses_per_epoch
def feature_metrix_contrastive(inputs, target, model, distance, criterion, opt, losses_per_epoch, reducer_pos, reducer_neg, alpha):
opt.zero_grad()
out_clean = model(inputs, 0.0, 0.0, 0.0, 0.0, mode='clean', type='clean')
loss_clean = criterion(out_clean, target)
a1, pos, a2, neg = lmu.get_all_pairs_indices(target)
layer_mix = random.randint(0,1)
if (layer_mix == 0):
lam_pn = np.random.beta(alpha, alpha)
new_a2, new_pos, new_neg = feature_posneg_pair_mixup_for_pos_anchor(a1, pos, a2, neg)
y_pos = target[new_pos.long()]
y_neg = target[new_neg.long()]
anchor_embedding = model(inputs, new_a2, new_pos, new_neg, lam_pn, mode='pos_neg_mixup', type='clean_anchor')
mixed_embedding = model(inputs, new_a2, new_pos, new_neg, lam_pn, mode='pos_neg_mixup', type='mixed')
sim_mix = distance(anchor_embedding, mixed_embedding)
loss_pos_dict, loss_neg_dict = loss_posneg_mixup(sim_mix, new_pos, new_a2, new_neg, lam_pn)
loss_mixed_posneg_for_pos_anc = reducer_pos(loss_pos_dict, sim_mix, y_pos) + reducer_neg(loss_neg_dict, sim_mix, y_neg)
loss = loss_clean + 0.4*loss_mixed_posneg_for_pos_anc
elif (layer_mix == 1):
lam_an = np.random.beta(alpha, alpha)
new_a2_an, new_neg_an = feature_neg_pair_mixup_without_posanchor(a1, a2, neg)
y_pos_an = target[new_a2_an.long()]
y_neg_an = target[new_neg_an.long()]
anchor_embedding_an = model(inputs, new_a2_an, new_a2_an, new_neg_an, lam_an, mode='anc_neg_mixup', type='clean_anchor')
mixed_embedding_an = model(inputs, new_a2_an, new_a2_an, new_neg_an, lam_an, mode='anc_neg_mixup', type='mixed')
sim_mix_an = distance(anchor_embedding_an, mixed_embedding_an)
loss_pos_dict_an, loss_neg_dict_an = loss_posneg_mixup(sim_mix_an, new_a2_an, new_a2_an, new_neg_an, lam_an)
loss_mixed_ancneg_without_pos_anc = reducer_pos(loss_pos_dict_an, sim_mix_an, y_pos_an) + reducer_neg(loss_neg_dict_an, sim_mix_an, y_neg_an)
loss = loss_clean + 0.4*loss_mixed_ancneg_without_pos_anc
loss.backward()
torch.nn.utils.clip_grad_value_(model.parameters(), 10)
losses_per_epoch.append(loss.data.cpu().numpy())
opt.step()
return loss, losses_per_epoch