-
Notifications
You must be signed in to change notification settings - Fork 2
/
utils.py
302 lines (240 loc) · 11 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
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
import re
import pandas as pd
from pathlib import Path
from collections import defaultdict
from sklearn.ensemble import RandomForestClassifier
import torch
import torchvision.transforms as transforms
from torch.utils.data.sampler import SubsetRandomSampler
import moco.loader
import moco.builder
from datasets import *
from models import *
# -------------------------- #
# dataset #
# -------------------------- #
def get_sup_dataloader(modelname, datapath, year, batchsize, workers, sequencelength, num, interp, rc,
useall=False, nclasses=20, seed=111):
train_dataaug = RandomTempShift()
if modelname in ['rf', 'RF']:
num_train = int(num * 0.9)
num_val = num - num_train
traindataset = USCrops(mode='train', root=datapath, year=year, sequencelength=sequencelength,
dataaug=train_dataaug, num=num_train, interp=interp, nclasses=nclasses, seed=seed)
testdataset = USCrops(mode='eval', root=datapath, year=year, sequencelength=sequencelength,
useall=useall, num=num_val, nclasses=nclasses, interp=interp, seed=seed)
X_train = traindataset.X_list
for i, X in enumerate(X_train):
X_train[i] = USCrops.transform(traindataset, X, interp=interp, rc=rc)[0].numpy()
X_train = np.array(X_train).reshape(len(X_train), -1)
y_train = traindataset.index['classid'].values
X_test = testdataset.X_list
for i, X in enumerate(X_test):
X_test[i] = USCrops.transform(testdataset, X, interp=interp, rc=rc)[0].numpy()
X_test = np.array(X_test).reshape(len(X_test), -1)
y_test = testdataset.index['classid'].values
meta = dict(
ndims=10*sequencelength,
num_classes=traindataset.nclasses+1,
)
return (X_train, y_train, X_test, y_test), meta
else:
num_train = int(num * 0.9)
num_val = num - num_train
traindataset = USCrops(mode='train', root=datapath, year=year, sequencelength=sequencelength, dataaug=train_dataaug,
useall=useall, num=num_train, randomchoice=rc, interp=interp, nclasses=nclasses, seed=seed)
valdataset = USCrops(mode='valid', root=datapath, year=year, sequencelength=sequencelength,
useall=useall, num=num_val, randomchoice=rc, interp=interp, nclasses=nclasses, seed=seed)
testdataset = USCrops(mode='eval', root=datapath, year=year, sequencelength=sequencelength,
useall=useall, num=num_val, randomchoice=rc, interp=interp, nclasses=nclasses, seed=seed)
traindataloader = torch.utils.data.DataLoader(traindataset, batch_size=batchsize, shuffle=True,
num_workers=workers, pin_memory=True)
valdataloader = torch.utils.data.DataLoader(valdataset, batch_size=batchsize, shuffle=False,
num_workers=workers)
testdataloader = torch.utils.data.DataLoader(testdataset, batch_size=batchsize, shuffle=False,
num_workers=workers, pin_memory=True)
meta = dict(
ndims=10,
num_classes=traindataset.nclasses+1,
)
return (traindataloader, valdataloader, testdataloader), meta
def get_moco_dataloader(datapath, year, batchsize, workers, sequencelength, num, rc, seed, useall):
train_dataaug = transforms.Compose([
RandomTempShift(),
RandomAddNoise(),
RandomTempRemoval(),
RandomSampleTimeSteps(sequencelength, rc=rc)
])
pretraindataset = MoCoDataset(root=datapath, year=year, sequencelength=sequencelength,
dataaug=train_dataaug, num=num, randomchoice=rc, seed=seed, useall=useall)
num = len(pretraindataset)
num_train = int(num * 0.9)
indices = list(range(num))
np.random.shuffle(indices)
train_idx, valid_idx = indices[:num_train], indices[num_train:]
train_sampler = SubsetRandomSampler(train_idx)
valid_sampler = SubsetRandomSampler(valid_idx)
traindataloader = torch.utils.data.DataLoader(pretraindataset, batch_size=batchsize, sampler=train_sampler,
num_workers=workers, pin_memory=True, drop_last=True)
valdataloader = torch.utils.data.DataLoader(pretraindataset, batch_size=batchsize, sampler=valid_sampler,
num_workers=workers, drop_last=True)
meta = dict(
ndims=10,
)
return traindataloader, valdataloader, meta
def get_bert_dataloader(datapath, year, batchsize, workers, sequencelength, num, rc, seed, useall):
pretraindataset = BERTDataset(root=datapath, year=year, sequencelength=sequencelength,
num=num, randomchoice=rc, seed=seed, useall=useall)
num = len(pretraindataset)
num_train = int(num * 0.9)
indices = list(range(num))
np.random.shuffle(indices)
train_idx, valid_idx = indices[:num_train], indices[num_train:]
train_sampler = SubsetRandomSampler(train_idx)
valid_sampler = SubsetRandomSampler(valid_idx)
traindataloader = torch.utils.data.DataLoader(pretraindataset, batch_size=batchsize, sampler=train_sampler,
num_workers=workers, pin_memory=True, drop_last=True)
valdataloader = torch.utils.data.DataLoader(pretraindataset, batch_size=batchsize, sampler=valid_sampler,
num_workers=workers, drop_last=True)
meta = dict(
ndims=10,
)
return traindataloader, valdataloader, meta
# -------------------------- #
# Model #
# -------------------------- #
def get_model(modelname, ndims, num_classes, sequencelength, device):
modelname = modelname.lower() # make case invariant
if modelname == 'transformer':
model = TransformerModel(input_dim=ndims, num_classes=num_classes, max_seq_len=sequencelength).to(device)
elif modelname == 'tempcnn':
model = TempCNN(input_dim=ndims, num_classes=num_classes, max_seq_len=sequencelength).to(device)
elif modelname == 'lstm':
model = LSTM(input_dim=ndims, num_classes=num_classes).to(device)
elif modelname == 'ltae':
model = LTAE(input_dim=ndims, num_classes=num_classes, max_seq_len=sequencelength).to(device)
elif modelname == 'rf':
model = RandomForestClassifier(n_estimators=500, max_depth=25)
elif modelname == 'stnet':
model = STNet(input_dim=ndims, num_classes=num_classes, max_seq_len=sequencelength).to(device)
else:
raise ValueError(
"invalid model argument. choose from 'Transformer', 'TempCNN', 'LSTM', 'LTAE', 'RF', or 'STNet' ")
return model
def get_moco_model(modelname, device, args):
modelname = modelname.lower()
if modelname == 'transformer':
basemodel = TransformerModel
elif modelname == 'tempcnn':
basemodel = TempCNN
elif modelname == 'lstm':
basemodel = LSTM
elif modelname == 'ltae':
basemodel = LTAE
elif modelname == 'stnet':
basemodel = STNet
else:
raise ValueError(
"invalid model - basemodel argument")
model = moco.builder.MoCo(
basemodel,
args.moco_dim, args.moco_k, args.moco_m, args.moco_t, args.mlp)
model.modelname = f'{model.modelname}{basemodel().modelname}'
model = model.to(device)
return model
# -------------------------- #
# Utils #
# -------------------------- #
class AverageMeter(object):
"""Computes and stores the average and current value"""
def __init__(self, name, fmt=':f'):
self.name = name
self.fmt = fmt
self.reset()
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
def __str__(self):
fmtstr = '{name} {val' + self.fmt + '} ({avg' + self.fmt + '})'
return fmtstr.format(**self.__dict__)
def accuracy(output, target, num_classes=21):
num = target.shape[0]
confusion_matrix = get_confusion_matrix(output, target, num_classes)
TP = confusion_matrix.diagonal()
FP = confusion_matrix.sum(1) - TP
FN = confusion_matrix.sum(0) - TP
po = TP.sum() / num
pe = (confusion_matrix.sum(0) * confusion_matrix.sum(1)).sum() / num ** 2
if pe == 1:
kappa = 1
else:
kappa = (po - pe) / (1 - pe)
p = TP / (TP + FP + 1e-12)
r = TP / (TP + FN + 1e-12)
f1 = 2 * p * r / (p + r + 1e-12)
oa = po
kappa = kappa
macro_f1 = f1.mean()
weight = confusion_matrix.sum(0) / confusion_matrix.sum()
weighted_f1 = (weight * f1).sum()
class_f1 = f1
return dict(
oa=oa,
kappa=kappa,
macro_f1=macro_f1,
weighted_f1=weighted_f1,
class_f1=class_f1,
confusion_matrix=confusion_matrix
)
def get_confusion_matrix(y_pred, y_true, num_classes=21):
idx = y_pred * num_classes + y_true
return np.bincount(idx, minlength=num_classes * num_classes).reshape(num_classes, num_classes)
def get_ntrainparams(model):
return sum(p.numel() for p in model.parameters() if p.requires_grad)
def adjust_learning_rate(optimizer, epoch, args):
"""Decay the learning rate based on schedule"""
lr = args.learning_rate
for milestone in args.schedule:
lr *= 0.1 if epoch >= milestone else 1.
for param_group in optimizer.param_groups:
param_group['lr'] = lr
def recursive_todevice(x, device):
if isinstance(x, torch.Tensor):
return x.to(device)
else:
return [recursive_todevice(c, device) for c in x]
def save(model, path="model.pth", **kwargs):
print(f"saving model to {str(path)}\n")
model_state = model.state_dict()
Path(path).parent.mkdir(exist_ok=True, parents=True)
torch.save(dict(model_state=model_state, **kwargs), path)
def overall_performance(logdir):
overall_metrics = defaultdict(list)
for seed in [111, 222, 333, 444, 555]:
log_dir = Path(logdir.replace(re.findall('Seed\d+', str(logdir))[0], f'Seed{seed}'))
log_fn = log_dir / f'testlog.csv'
if log_fn.exists():
test_metrics = pd.read_csv(log_fn).iloc[0].to_dict()
for metric, value in test_metrics.items():
overall_metrics[metric].append(value)
print(f'Overall result across 5 trials:')
for metric, values in overall_metrics.items():
values = np.array(values)
if isinstance(values[0], (str)) or np.any(np.isnan(values)):
continue
if 'loss' in metric or 'f1' in metric or 'kappa' in metric:
print(f"{metric}: {np.mean(values):.4}")
else:
values *= 100
print(f"{metric}: {np.mean(values):.2f}")
print(f'{np.mean(overall_metrics["oa"])*100:.2f}\t'
f'{np.mean(overall_metrics["kappa"]):.4f}\t'
f'{np.mean(overall_metrics["weighted_f1"]):.4f}')
print()