-
Notifications
You must be signed in to change notification settings - Fork 1
/
utils.py
223 lines (174 loc) · 8.34 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
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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
# current implementation: only support numerical values
import numpy as np
import torch, os
from torch import nn as nn
from torch.utils.data import Dataset, DataLoader
import pandas as pd
import math
import argparse
class MaskEmbed(nn.Module):
""" record to mask embedding
"""
def __init__(self, rec_len=25, embed_dim=64, norm_layer=None):
super().__init__()
self.rec_len = rec_len
self.proj = nn.Conv1d(1, embed_dim, kernel_size=1, stride=1)
self.norm = norm_layer(embed_dim) if norm_layer else nn.Identity()
def forward(self, x):
B, _, L = x.shape
# assert(L == self.rec_len, f"Input data width ({L}) doesn't match model ({self.rec_len}).")
x = self.proj(x)
x = x.transpose(1, 2)
x = self.norm(x)
return x
class ActiveEmbed(nn.Module):
""" record to mask embedding
"""
def __init__(self, rec_len=25, embed_dim=64, norm_layer=None):
super().__init__()
self.rec_len = rec_len
self.proj = nn.Conv1d(1, embed_dim, kernel_size=1, stride=1)
self.norm = norm_layer(embed_dim) if norm_layer else nn.Identity()
def forward(self, x):
B, _, L = x.shape
# assert(L == self.rec_len, f"Input data width ({L}) doesn't match model ({self.rec_len}).")
x = self.proj(x)
x = torch.sin(x)
x = x.transpose(1, 2)
# x = torch.cat((torch.sin(x), torch.cos(x + math.pi/2)), -1)
x = self.norm(x)
return x
def get_1d_sincos_pos_embed(embed_dim, pos, cls_token=False):
"""
embed_dim: output dimension for each position
pos: a list of positions to be encoded: size (M,)
out: (M, D)
"""
assert embed_dim % 2 == 0
omega = np.arange(embed_dim // 2, dtype=np.float)
omega /= embed_dim / 2.
omega = 1. / 10000**omega # (D/2,)
pos = np.arange(pos) # (M,)
out = np.einsum('m,d->md', pos, omega) # (M, D/2), outer product
emb_sin = np.sin(out) # (M, D/2)
emb_cos = np.cos(out) # (M, D/2)
pos_embed = np.concatenate([emb_sin, emb_cos], axis=1) # (M, D)
if cls_token:
pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
return pos_embed
def adjust_learning_rate(optimizer, epoch, lr, min_lr, max_epochs, warmup_epochs):
"""Decay the learning rate with half-cycle cosine after warmup"""
if epoch < warmup_epochs:
tmp_lr = lr * epoch / warmup_epochs
else:
tmp_lr = min_lr + (lr - min_lr) * 0.5 * \
(1. + math.cos(math.pi * (epoch - warmup_epochs) / (max_epochs - warmup_epochs)))
for param_group in optimizer.param_groups:
if "lr_scale" in param_group:
param_group["lr"] = tmp_lr * param_group["lr_scale"]
else:
param_group["lr"] = tmp_lr
return tmp_lr
def get_grad_norm_(parameters, norm_type: float = 2.0) -> torch.Tensor:
if isinstance(parameters, torch.Tensor):
parameters = [parameters]
parameters = [p for p in parameters if p.grad is not None]
norm_type = float(norm_type)
if len(parameters) == 0:
return torch.tensor(0.)
device = parameters[0].grad.device
if norm_type == np.inf:
total_norm = max(p.grad.detach().abs().max().to(device) for p in parameters)
else:
total_norm = torch.norm(torch.stack([torch.norm(p.grad.detach(), norm_type).to(device) for p in parameters]), norm_type)
return total_norm
class NativeScaler:
state_dict_key = "amp_scaler"
def __init__(self):
self._scaler = torch.cuda.amp.GradScaler()
def __call__(self, loss, optimizer, clip_grad=None, parameters=None, create_graph=False, update_grad=True):
self._scaler.scale(loss).backward(create_graph=create_graph)
if update_grad:
if clip_grad is not None:
assert parameters is not None
self._scaler.unscale_(optimizer) # unscale the gradients of optimizer's assigned params in-place
norm = torch.nn.utils.clip_grad_norm_(parameters, clip_grad)
else:
self._scaler.unscale_(optimizer)
norm = get_grad_norm_(parameters)
self._scaler.step(optimizer)
self._scaler.update()
else:
norm = None
return norm
def state_dict(self):
return self._scaler.state_dict()
def load_state_dict(self, state_dict):
self._scaler.load_state_dict(state_dict)
class MAEDataset(Dataset):
def __init__(self, X, M):
self.X = X
self.M = M
def __len__(self):
return len(self.X)
def __getitem__(self, idx: int):
return self.X[idx], self.M[idx]
def get_dataset(dataset : str, path : str):
if dataset in ['climate', 'compression', 'wine', 'yacht', 'spam', 'letter', 'credit', 'raisin', 'bike', 'obesity', 'airfoil', 'blood', 'yeast', 'health', 'review', 'travel']:
df = pd.read_csv(os.path.join(path, 'data', dataset + '.csv'))
last_col = df.columns[-1]
y = df[last_col]
X = df.drop(columns=[last_col])
elif dataset == 'california':
from sklearn.datasets import fetch_california_housing
X, y = fetch_california_housing(as_frame=True, return_X_y=True)
elif dataset == 'diabetes':
from sklearn.datasets import load_diabetes
X, y = load_diabetes(as_frame=True, return_X_y=True)
elif dataset == 'iris':
# only for testing
from sklearn.datasets import load_iris
X, y = load_iris(as_frame=True, return_X_y=True)
return X, y
def get_args_parser():
parser = argparse.ArgumentParser('MAE pre-training', add_help=False)
parser.add_argument('--dataset', default='california', type=str)
parser.add_argument('--batch_size', default=64, type=int,
help='Batch size per GPU (effective batch size is batch_size * accum_iter * # gpus')
parser.add_argument('--max_epochs', default=600, type=int)
parser.add_argument('--accum_iter', default=1, type=int,
help='Accumulate gradient iterations (for increasing the effective batch size under memory constraints)')
# Model parameters
parser.add_argument('--mask_ratio', default=0.5, type=float, help='Masking ratio (percentage of removed patches).')
parser.add_argument('--embed_dim', default=32, type=int, help='embedding dimensions')
parser.add_argument('--depth', default=6, type=int, help='encoder depth')
parser.add_argument('--decoder_depth', default=4, type=int, help='decoder depth')
parser.add_argument('--num_heads', default=4, type=int, help='number of heads')
parser.add_argument('--mlp_ratio', default=4., type=float, help='mlp ratio')
parser.add_argument('--encode_func', default='linear', type=str, help='encoding function')
parser.add_argument('--norm_field_loss', default=False,
help='Use (per-patch) normalized field as targets for computing loss')
parser.set_defaults(norm_field_loss=False)
# Optimizer parameters
parser.add_argument('--weight_decay', type=float, default=0.05, help='weight decay (default: 0.05)')
parser.add_argument('--lr', type=float, default=None, metavar='LR', help='learning rate (absolute lr)')
parser.add_argument('--blr', type=float, default=1e-3, metavar='LR',
help='base learning rate: absolute_lr = base_lr * total_batch_size / 256')
parser.add_argument('--min_lr', type=float, default=1e-5, metavar='LR',
help='lower lr bound for cyclic schedulers that hit 0')
parser.add_argument('--warmup_epochs', type=int, default=40, metavar='N', help='epochs to warmup LR')
###### change this path
parser.add_argument('--path', default='/data/tianyu/remasker/', type=str, help='dataset path')
parser.add_argument('--exp_name', default='test', type=str, help='experiment name')
# Dataset parameters
parser.add_argument('--device', default='cuda', help='device to use for training / testing')
parser.add_argument('--seed', default=666, type=int)
parser.add_argument('--overwrite', default=True, help='whether to overwrite default config')
parser.add_argument('--pin_mem', action='store_false')
# distributed training parameters
return parser
if __name__ == '__main__':
X = torch.tensor([[1, 2, 3, 4]], dtype=torch.float32)
X = X.unsqueeze(1)
mask_embed = ActiveEmbed(4, 6)
print(mask_embed(X).shape)