-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
139 lines (117 loc) · 4.62 KB
/
utils.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
import json
import math
import random
import numpy as np
import torch
def set_seed(seed):
""" set the random seed """
seed = int(seed)
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
def chunks(arr, m):
""" split the arr into N chunks. """
return [arr[i:i + n] for i in range(0, len(arr), n)]
def get_curriculum_stages(y, s, idx_path, N=2):
""" split the index into N shares in origin order.
Arguments:
idx_path(str): the index storage path
N(int): the number of stages
Return:
"""
with open(idx_path) as f:
idx = json.load(f)
idx = np.array(idx)
y_sorted = y[idx]
s_sorted = s[idx]
classes = np.unique(y)
s_feat = np.unique(s)
# N = 2 时,就是每个 subgroup 按顺序取一半,放在一起
# step1: get the subgroup: {(g0, y0): [...]}
all_grp = {}
# step2: calculate each group size in one stage
grp_size = {}
# y
for g in s_feat:
for label in classes:
# order messy?
sub_grp = idx[(s_sorted == g) & (y_sorted == label)]
all_grp[(g, label)] = sub_grp
grp_size[(g, label)] = len(sub_grp) / N
res = []
for i in range(N):
tmp = []
print('=' * 20)
for g, label in all_grp.keys():
size = int(grp_size[(g, label)])
sub_grp = all_grp[(g, label)]
j = i * size
print('g:%d, y:%d ==> size:%d' % (g, label, size))
tmp.append(sub_grp[j: j + size])
res.append(np.concatenate(tmp))
return res
class EarlyStopping:
"""Early stops the training if validation loss doesn't improve after a given patience."""
def __init__(self, patience=7, verbose=False, delta=0, root_path='checkpoint/', trace_func=print):
"""
Args:
patience (int): How long to wait after last time validation loss improved.
Default: 7
verbose (bool): If True, prints a message for each validation loss improvement.
Default: False
delta (float): Minimum change in the monitored quantity to qualify as an improvement.
Default: 0
path (str): Path for the checkpoint to be saved to.
Default: 'checkpoint.pt'
trace_func (function): trace print function.
Default: print
"""
self.patience = patience
self.verbose = verbose
self.counter = 0
self.best_score = None
self.early_stop = False
self.val_loss_min = np.Inf
self.delta = delta
self.root_path = root_path
self.trace_func = trace_func
def __call__(self, val_loss, model=None):
score = -val_loss
if self.best_score is None:
self.best_score = score
if model is not None:
self.save_checkpoint(val_loss, model)
elif score < self.best_score + self.delta:
self.counter += 1
self.trace_func(f'EarlyStopping counter: {self.counter} out of {self.patience}')
if self.counter >= self.patience:
self.early_stop = True
else:
self.best_score = score
if model is not None:
self.save_checkpoint(val_loss, model)
self.counter = 0
def save_checkpoint(self, val_loss, model):
'''Saves model when validation loss decrease.'''
if self.verbose:
self.trace_func(
f'Validation loss decreased ({self.val_loss_min:.6f} --> {val_loss:.6f}). Saving model ...')
file_path = self.root_path + model.scope_name + str(model.batch_id) + '.pt'
if model.debias:
save_dict = {
'model_clf_model_state_dict': model.clf_model.state_dict(),
'model_clf_opt_state_dict': model.classifier_opt.state_dict(),
'model_clf_lr_scheduler_state_dict': model.clf_lr_scheduler.state_dict(),
'model_adv_model_state_dict': model.adv_model.state_dict(),
'model_adversary_opt_state_dict': model.adversary_opt.state_dict(),
'model_adv_lr_scheduler': model.adv_lr_scheduler.state_dict()
}
else:
save_dict = {
'model_clf_model_state_dict': model.clf_model.state_dict(),
'model_clf_opt_state_dict': model.classifier_opt.state_dict(),
'model_clf_lr_scheduler_state_dict': model.clf_lr_scheduler.state_dict()
}
torch.save(save_dict, file_path)
self.val_loss_min = val_loss