-
Notifications
You must be signed in to change notification settings - Fork 3
/
comet.py
306 lines (248 loc) · 12.1 KB
/
comet.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
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
import torch
import torch.nn.functional as F
from torch import nn
from torch.utils.data import TensorDataset, DataLoader
import numpy as np
from models import TSEncoder, ProjectionHead
from models.losses import contrastive_loss
from models.losses import sample_contrastive_loss, observation_contrastive_loss, patient_contrastive_loss, trial_contrastive_loss
from utils import batch_shuffle_feature_label, trial_shuffle_feature_label, shuffle_feature_label
from utils import MyBatchSampler
import math
import copy
import sklearn
class COMET:
"""The COMET model"""
def __init__(
self,
input_dims,
output_dims=320,
hidden_dims=64,
depth=10,
device='cuda',
lr=0.001,
batch_size=128,
after_iter_callback=None,
after_epoch_callback=None,
flag_use_multi_gpu=True
):
""" Initialize a COMET model.
Args:
input_dims (int): The input dimension. For a uni-variate time series, this should be set to 1.
output_dims (int): The representation dimension.
hidden_dims (int): The hidden dimension of the encoder.
depth (int): The number of hidden residual blocks in the encoder.
device (str): The gpu used for training and inference.
lr (float): The learning rate.
batch_size (int): The batch size of samples.
after_iter_callback (Union[Callable, NoneType]): A callback function that would be called after each iteration.
after_epoch_callback (Union[Callable, NoneType]): A callback function that would be called after each epoch.
flag_use_multi_gpu (bool): A flag to indicate whether using multiple gpus
"""
super().__init__()
self.device = device
self.lr = lr
self.batch_size = batch_size
self.output_dims = output_dims
self.hidden_dims = hidden_dims
self.flag_use_multi_gpu = flag_use_multi_gpu
# gpu_idx_list = [0, 1]
self._net = TSEncoder(input_dims=input_dims, output_dims=output_dims, hidden_dims=hidden_dims, depth=depth)
device = torch.device(device)
if device == torch.device('cuda') and self.flag_use_multi_gpu:
# self._net = nn.DataParallel(self._net, device_ids=gpu_idx_list)
self._net = nn.DataParallel(self._net)
self._net.to(device)
# stochastic weight averaging
# https://pytorch.org/blog/pytorch-1.6-now-includes-stochastic-weight-averaging/
# self.net = self._net
self.net = torch.optim.swa_utils.AveragedModel(self._net)
self.net.update_parameters(self._net)
# projection head append after encoder
# self.proj_head = ProjectionHead(input_dims=self.output_dims, output_dims=2, hidden_dims=128).to(self.device)
self.after_iter_callback = after_iter_callback
self.after_epoch_callback = after_epoch_callback
self.n_epochs = 1
self.n_iters = 1
def fit(self, X, y, shuffle_function='trial', masks=None, factors=None, n_epochs=None, n_iters=None, verbose=True):
""" Training the COMET model.
Args:
X (numpy.ndarray): The training data. It should have a shape of (n_samples, sample_timestamps, features).
y (numpy.ndarray): The training labels. It should have a shape of (n_samples, 3). The three columns are the label, patient id, and trial id.
shuffle_function (str): specify the shuffle function.
masks (list): A list of masking functions applied (str). [Patient, Trial, Sample, Observation].
factors (list): A list of loss factors. [Patient, Trial, Sample, Observation].
n_epochs (Union[int, NoneType]): The number of epochs. When this reaches, the training stops.
n_iters (Union[int, NoneType]): The number of iterations. When this reaches, the training stops. If both n_epochs and n_iters are not specified, a default setting would be used that sets n_iters to 200 for a dataset with size <= 100000, 600 otherwise.
verbose (bool): Whether to print the training loss after each epoch.
Returns:
epoch_loss_list: a list containing the training losses on each epoch.
epoch_f1_list: a list containing the linear evaluation on validation f1 score on each epoch.
"""
assert X.ndim == 3
assert y.shape[1] == 3
# Important!!! Shuffle the training set for contrastive learning pretraining. Check details in utils.py.
X, y = shuffle_feature_label(X, y, shuffle_function=shuffle_function, batch_size=self.batch_size)
if n_iters is None and n_epochs is None:
n_iters = 200 if X.size <= 100000 else 600 # default param for n_iters
# we need patient id for patient-level contrasting and trial id for trial-level contrasting
train_dataset = TensorDataset(torch.from_numpy(X).to(torch.float), torch.from_numpy(y).to(torch.float))
if shuffle_function == 'random':
train_loader = DataLoader(train_dataset, batch_size=min(self.batch_size, len(train_dataset)), shuffle=True,
drop_last=False)
else:
# Important!!! A customized batch_sampler to shuffle samples before each epoch. Check details in utils.py.
my_sampler = MyBatchSampler(range(len(train_dataset)), batch_size=min(self.batch_size, len(train_dataset)), drop_last=False)
train_loader = DataLoader(train_dataset, batch_sampler=my_sampler)
optimizer = torch.optim.AdamW(self._net.parameters(), lr=self.lr)
epoch_loss_list, epoch_f1_list = [], []
while True:
# count by epoch
if n_epochs is not None and self.n_epochs >= n_epochs:
break
cum_loss = 0
n_epoch_iters = 1
interrupted = False
for x, y in train_loader:
# count by iterations
if n_iters is not None and self.n_iters >= n_iters:
interrupted = True
break
x = x.to(self.device)
pid = y[:, 1].to(self.device) # patient id
tid = y[:, 2].to(self.device) # trial id
optimizer.zero_grad()
# positive pairs construction
# the fixed random seed guarantee reproducible
# but does not mean the same mask will generate same result in one running
if masks is None:
masks = ['all_true', 'all_true', 'continuous', 'continuous']
if factors is None:
factors = [0.25, 0.25, 0.25, 0.25]
if factors[0] != 0:
# do augmentation and compute representation
patient_out1 = self._net(x, mask=masks[0])
patient_out2 = self._net(x, mask=masks[0])
# loss calculation
patient_loss = contrastive_loss(
patient_out1,
patient_out2,
patient_contrastive_loss,
id=pid,
hierarchical=False,
factor=factors[0],
)
else:
patient_loss = 0
if factors[1] != 0:
trial_out1 = self._net(x, mask=masks[1])
trial_out2 = self._net(x, mask=masks[1])
trial_loss = contrastive_loss(
trial_out1,
trial_out2,
trial_contrastive_loss,
id=tid,
hierarchical=False,
factor=factors[1],
)
else:
trial_loss = 0
if factors[2] != 0:
sample_out1 = self._net(x, mask=masks[2])
sample_out2 = self._net(x, mask=masks[2])
sample_loss = contrastive_loss(
sample_out1,
sample_out2,
sample_contrastive_loss,
hierarchical=True,
factor=factors[2],
)
else:
sample_loss = 0
if factors[3] != 0:
observation_out1 = self._net(x, mask=masks[3])
observation_out2 = self._net(x, mask=masks[3])
observation_loss = contrastive_loss(
observation_out1,
observation_out2,
observation_contrastive_loss,
hierarchical=True,
factor=factors[3],
)
else:
observation_loss = 0
loss = patient_loss + trial_loss + sample_loss + observation_loss
loss.backward()
optimizer.step()
self.net.update_parameters(self._net)
cum_loss += loss.item()
if self.after_iter_callback is not None:
self.after_iter_callback(self, loss.item())
n_epoch_iters += 1
self.n_iters += 1
if interrupted:
break
cum_loss /= n_epoch_iters
epoch_loss_list.append(cum_loss)
if verbose:
print(f"Epoch #{self.n_epochs}: loss={cum_loss}")
if self.after_epoch_callback is not None:
linear_f1 = self.after_epoch_callback(self, cum_loss)
epoch_f1_list.append(linear_f1)
self.n_epochs += 1
return epoch_loss_list, epoch_f1_list
def eval_with_pooling(self, x, mask=None):
""" Pooling the representation.
"""
out = self.net(x, mask)
# representation shape: B x O x Co ---> B x 1 x Co ---> B x Co
out = F.max_pool1d(
out.transpose(1, 2),
kernel_size=out.size(1),
).transpose(1, 2)
out = out.squeeze(1)
return out
def encode(self, X, mask=None, batch_size=None):
""" Compute representations using the model.
Args:
X (numpy.ndarray): The input data. This should have a shape of (n_samples, sample_timestamps, features).
mask (str): The mask used by encoder can be specified with this parameter. Check masking functions in encoder.py.
batch_size (Union[int, NoneType]): The batch size used for inference. If not specified, this would be the same batch size as training.
Returns:
repr: The representations for data.
"""
assert self.net is not None, 'please train or load a net first'
assert X.ndim == 3
if batch_size is None:
batch_size = self.batch_size
# n_samples, ts_l, _ = data.shape
org_training = self.net.training
self.net.eval()
dataset = TensorDataset(torch.from_numpy(X).to(torch.float))
loader = DataLoader(dataset, batch_size=batch_size)
with torch.no_grad():
output = []
for batch in loader:
x = batch[0].to(self.device)
# print(next(self.net.parameters()).device)
# print(x.device)
out = self.eval_with_pooling(x, mask)
output.append(out)
output = torch.cat(output, dim=0)
self.net.train(org_training)
# return output.numpy()
return output.cpu().numpy()
def save(self, fn):
""" Save the model to a file.
Args:
fn (str): filename.
"""
torch.save(self.net.state_dict(), fn)
def load(self, fn):
""" Load the model from a file.
Args:
fn (str): filename.
"""
# state_dict = torch.load(fn, map_location=self.device)
state_dict = torch.load(fn)
self.net.load_state_dict(state_dict)