-
Notifications
You must be signed in to change notification settings - Fork 1
/
validate.py
441 lines (403 loc) · 21.8 KB
/
validate.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
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
import torch
import torch.nn as nn
import numpy as np
import torch.nn.functional as F
import os
import yaml
from torch.utils.data.dataloader import DataLoader
from tqdm import tqdm
from net.CombinatorialNetwork import CombinatorialNet
from net.confidence import confidenceNetwork
from newdataset import gridNewDataset
#注意，这是validate，不是最终生成的test
class Validate(nn.Module):
def __init__(self, combinatorial_args, data_iter, device):
super(Validate, self).__init__()
self.confidence = confidenceNetwork()
self.prediction = CombinatorialNet(
combinatorial_args['encoder']['in_channels'],
combinatorial_args['encoder']['mid_channels'],
combinatorial_args['encoder']['out_channels'],
combinatorial_args['ordinal']['mid_channels'],
combinatorial_args['ordinal']['out_channels'],
combinatorial_args['decoder']['mid_channels'],
combinatorial_args['decoder']['out_channels'],
combinatorial_args['nclass'],
noise_mean=0,
noise_std=1e-1)
self.data_iter = data_iter
self.device = device
self.nClass = combinatorial_args['nclass']
def initialize(self, confidence_path, encoder_path, decoder_path,
ordinal_path):
confidence_ckpt = torch.load(confidence_path)
encoder_ckpt = torch.load(encoder_path)
decoder_ckpt = torch.load(decoder_path)
ordinal_ckpt = torch.load(ordinal_path)
self.confidence.load_state_dict(confidence_ckpt)
self.confidence.eval()
self.prediction.encoder.load_state_dict(encoder_ckpt)
self.prediction.decoder.load_state_dict(decoder_ckpt)
self.prediction.OD.load_state_dict(ordinal_ckpt)
self.prediction.eval()
def forward(self):
tsthreas = [0.1, 3, 10, 20]
tp = [0] * len(tsthreas) # true positive
tn = [0] * len(tsthreas) # true negetive
fp = [0] * len(tsthreas) # false positve
fn = [0] * len(tsthreas) # false negetive
ts = [0] * len(tsthreas)
totalRegressionLoss = [] #prediction loss
totalReconstructLoss = []
totalClassificationLoss = [] #传统regression loss
totalRClassificationLoss = [] #confidence网络 loss
total_error = 0
total_count = 0
p_error = 0
ps_error = 0
p_count = 0
pxErrorList = [0] * (self.nClass)
pxsErrorList = [0] * (self.nClass)
pxCountList = [0] * (self.nClass)
pxAverageError = [0] * (self.nClass)
pxsAverageError = [0] * (self.nClass)
classCorrect = [0] * (self.nClass)
classCounnt = [0] * (self.nClass)
accuray = [0] * (self.nClass + 1)
rainCorrect = [0] * 2
rainCount = [0] * 2
rainAccuracy = [0] * 3
for i, iter in enumerate(tqdm(self.data_iter, desc="validating: ")):
input, rain, temp, time = iter
input = input.type(torch.FloatTensor).to(self.device)
rain = rain.type(torch.FloatTensor).to(self.device) #(N, 1)
temp = temp.type(torch.FloatTensor).to(self.device)
#print('label:', rain, temp)
#TODO: 这里要对标签数据进行分类处理，得到rainClass
rainClass = torch.Tensor().to(self.device)
rainClass_1dim = torch.Tensor().to(self.device)
for j in range(rain.shape[0]):
#print(rain[j])
if rain[j] == -99999:
#TODO: 怎么处理-99999？
rainClass = torch.concat(
(rainClass, torch.Tensor([0, 0]).to(self.device)))
rainClass_1dim = torch.concat(
(rainClass_1dim, torch.Tensor([0]).to(self.device)))
elif rain[j] < 0.1:
rainClass = torch.concat(
(rainClass, torch.Tensor([1, 0]).to(self.device)))
rainClass_1dim = torch.concat(
(rainClass_1dim, torch.Tensor([0]).to(self.device)))
else:
rainClass = torch.concat(
(rainClass, torch.Tensor([0, 1]).to(self.device)))
rainClass_1dim = torch.concat(
(rainClass_1dim, torch.Tensor([1]).to(self.device)))
rainClass = rainClass.view(rain.shape[0], -1)
#print('rainClass:', rainClass, rainClass_1dim)
rainNumpy = rainClass_1dim.cpu().numpy()
gt_micaps = rain.cpu().numpy()
with torch.no_grad():
#三个网络分别的输出
reconstructValues = self.prediction(input, isOrdinal=False)
predictValues = self.prediction(
input, isOrdinal=True) #(N, 4==#classes)
rainPreds = self.confidence(input) #(N, 2)
#print('output ', predictValues, rainPreds)
#得到confidence网络的预测oneHot，是否下雨，还有个普通regressionPredict的方法预测是否下雨，暂时没用
rainPredsSoftMax = F.softmax(rainPreds, dim=1)
rainOnehot = self.generateOneHot(rainPredsSoftMax).to(
self.device)
#print('rainOnehot: ', rainOnehot, rainOnehot.shape)
#需要修改：看ordinal网络的输出
#TODO: 应该已经改过了？但我觉得不应该用0.5，因为这个应该是间隔？
#但感觉regressionValue的具体数值并不重要，只要分类对了就行
#REPO: 我也觉得这个0.5可以不用乘
regressionValues = 0.5 * (torch.sum(
(predictValues > 0.5).float(), dim=1).view(-1, 1)) #(N, 1)
zeros = torch.zeros(regressionValues.size()).to(self.device)
#print('regression:', regressionValues, regressionValues.shape)
#print('zeros:', zeros, zeros.shape)
# 将confidence网络的预测进行mask，得到最终的输出
#TODO: 感觉这种算法不对啊，除非rainOnehot是(N, 1, 2)，否则得不到(N,1,1)，但我觉得是(N,2)
#REPO: 实验下来，rainOnehot是(N,2),后一项是（N,2,1），乘出来是(N,N,1)[matmul的特性，后一项多出的第一维会作为batch，剩余二维部分做矩阵乘法]
regressionValues = torch.matmul(
rainOnehot, #(N, 2)
torch.cat((zeros, regressionValues),
dim=1).unsqueeze(-1)).squeeze(-1)
#print('regression new:', regressionValues,regressionValues.shape)
#regressionValues = regressionValues.item()
#计算重建损失和预测与标签的损失
regressionLoss = nn.MSELoss()(regressionValues, rain)
reconstructLoss = nn.MSELoss()(reconstructValues, input)
#TODO: 这里要用上分类处理后的标签数据
#计算confidence损失
rainClassificationLoss = nn.CrossEntropyLoss()(rainPreds,
rainClass)
#print('losses: ', regressionLoss, reconstructLoss,rainClassificationLoss)
#更新三个损失的总体list
totalRegressionLoss.append(regressionLoss.item())
totalReconstructLoss.append(reconstructLoss.item())
totalRClassificationLoss.append(rainClassificationLoss.item())
'''
#这是对普通regression做的
for i in range(self.nClass):
classCorrect[i] += np.sum(
(predicted == i) * (regressionNumpy == i) *
(rainNumpy == 1))
classCounnt[i] += np.sum(
(regressionNumpy == i) * (rainNumpy == 1))
'''
#计算confidence预测准确的个数
rainPredicted = torch.argmax(rainPredsSoftMax,
dim=1).cpu().numpy()
#print('rainPredicted', rainPredicted)
for i in range(2):
rainCorrect[i] += np.sum(
(rainPredicted == i) * (rainNumpy == i))
rainCount[i] += np.sum(rainNumpy == i)
#计算最终输出与GT的预测差距
predictNumpy = regressionValues.cpu().numpy()
gapValues = np.abs(predictNumpy - gt_micaps)
total_error += np.sum(gapValues)
total_count += gapValues.shape[0]
# 这里不知道输出的是啥,p_ae是一个gt中有降水（>=0.1的区域）的点的值为预测差距值的array
# p_error是有降水的点预测差的和
#TODO: 看起来应该是某些指标，具体公式可以后面再问问
p_ae = (gt_micaps >= 0.1) * gapValues
p_error += np.sum(p_ae)
ps_error += np.sum(p_ae**2)
p_count += np.sum(gt_micaps > 0.05)
'''
# 这是对传统regression做的
one_hot_mask = regressionMask.numpy()
for i in range(self.nClass):
ae = one_hot_mask[:, i].reshape(-1, 1) * gapValues
pxErrorList[i] += np.sum(ae)
pxsErrorList[i] += np.sum(ae**2)
pxCountList[i] += np.sum(one_hot_mask[:, i])'''
#TODO: 改成了直接用序回归出的分类来算，而不用预测出的具体降水值;
# 如果觉得不太合理就改回去吧
#REPO: 但这样的话，confidence网络就没用了，我们最终还是要给test生成具体降水值
# for i, threas in enumerate(tsthreas):
# tp[i] += np.sum(
# (gt_micaps >= threas) * (predictNumpy >= threas))
# tn[i] += np.sum(
# (gt_micaps < threas) * (predictNumpy < threas))
# fp[i] += np.sum(
# (gt_micaps < threas) * (predictNumpy >= threas))
# fn[i] += np.sum(
# (gt_micaps >= threas) * (predictNumpy < threas))
threshold_for_probability = 0.5
predictValues_numpy = predictValues.cpu().numpy()
#print(predictValues_numpy.shape)
for i, threas in enumerate(tsthreas):
tp[i] += np.sum(
(gt_micaps >= threas) * (predictValues_numpy[:, i] >=
threshold_for_probability))
tn[i] += np.sum(
(gt_micaps < threas) * (predictValues_numpy[:, i] <
threshold_for_probability))
fp[i] += np.sum(
(gt_micaps < threas) * (predictValues_numpy[:, i] >=
threshold_for_probability))
fn[i] += np.sum(
(gt_micaps >= threas) * (predictValues_numpy[:, i] <
threshold_for_probability))
#print('finals:', tp, tn, fp, fn)
#计算TS,四舍五入保留5位小数
for i, _ in enumerate(tsthreas):
ts[i] += round(tp[i] / (tp[i] + fp[i] + fn[i]), 5)
#计算总体平均误差——最终输出与GT预测的差/测试集大小
#p平均误差
#p平方后的平均误差
totalAverageError = round(total_error / total_count, 5)
pAverageError = round(p_error / p_count, 5)
psAverageError = round(ps_error / p_count - pAverageError**2, 5)
"""---------------- Total Loss for validation ----------------"""
totalLoss = np.mean(totalRegressionLoss)
totalRLoss = np.mean(totalReconstructLoss)
"""---------------- Total rainAccuracy, non-rainAccuracy and totalAccuracy for validation ----------------"""
for i in range(2):
rainAccuracy[i] += round(rainCorrect[i] / rainCount[i], 5)
rainAccuracy[2] += round(sum(rainCorrect) / sum(rainCount), 5)
# save TS Scores to list[(_,_,_,_,_)]
tsDisplay = list(zip(tp, tn, fp, fn, ts))
"""---------------- print validation info ----------------"""
info = {
"test_regression_loss": totalLoss,
"test_reconstruct_loss": totalRLoss,
"aver_gap": totalAverageError,
"aver_p_gap": pAverageError,
"aver_ps_gap": psAverageError,
"p_num": p_count,
"ts_score": tsDisplay,
"test_rain_classification_accuracy": rainAccuracy,
}
print(
"========================== Epoch {} Test Result Show =========================="
.format(1))
print(info)
def simple_validate(self):
tsthreas = [0.1, 3, 10, 20]
tp = [0] * len(tsthreas) # true positive
tn = [0] * len(tsthreas) # true negetive
fp = [0] * len(tsthreas) # false positve
fn = [0] * len(tsthreas) # false negetive
ts = [0] * len(tsthreas)
pxErrorList = [0] * (self.nClass)
pxsErrorList = [0] * (self.nClass)
pxCountList = [0] * (self.nClass)
pxAverageError = [0] * (self.nClass)
pxsAverageError = [0] * (self.nClass)
classCorrect = [0] * (self.nClass)
classCounnt = [0] * (self.nClass)
accuray = [0] * (self.nClass + 1)
rainCorrect = [0] * 2
rainCount = [0] * 2
rainAccuracy = [0] * 3
for i, iter in enumerate(tqdm(self.data_iter, desc="validating: ")):
input, rain, temp, time = iter
input = input.type(torch.FloatTensor).to(self.device)
rain = rain.type(torch.FloatTensor).to(self.device) #(N, 1)
temp = temp.type(torch.FloatTensor).to(self.device)
#print('label:', rain, temp)
#TODO: 这里要对标签数据进行分类处理，得到rainClass
rainClass = torch.Tensor().to(self.device)
rainClass_1dim = torch.Tensor().to(self.device)
for j in range(rain.shape[0]):
#print(rain[j])
if rain[j] == -99999:
#TODO: 怎么处理-99999？
rainClass = torch.concat(
(rainClass, torch.Tensor([0, 0]).to(self.device)))
rainClass_1dim = torch.concat(
(rainClass_1dim, torch.Tensor([0]).to(self.device)))
elif rain[j] < 0.1:
rainClass = torch.concat(
(rainClass, torch.Tensor([1, 0]).to(self.device)))
rainClass_1dim = torch.concat(
(rainClass_1dim, torch.Tensor([0]).to(self.device)))
else:
rainClass = torch.concat(
(rainClass, torch.Tensor([0, 1]).to(self.device)))
rainClass_1dim = torch.concat(
(rainClass_1dim, torch.Tensor([1]).to(self.device)))
rainClass = rainClass.view(rain.shape[0], -1)
#print('rainClass:', rainClass, rainClass_1dim)
rainNumpy = rainClass_1dim.cpu().numpy()
gt_micaps = rain.cpu().numpy()
with torch.no_grad():
#三个网络分别的输出
reconstructValues = self.prediction(input, isOrdinal=False)
predictValues = self.prediction(
input, isOrdinal=True) #(N, 4==#classes)
rainPreds = self.confidence(input) #(N, 2)
#print(torch.concat((predictValues, rain.unsqueeze(-1)), dim=1))
#得到confidence网络的预测oneHot，是否下雨，还有个普通regressionPredict的方法预测是否下雨，暂时没用
rainPredsSoftMax = F.softmax(rainPreds, dim=1)
rainOnehot = self.generateOneHot(rainPredsSoftMax).to(
self.device)
#print('rainOnehot: ', rainOnehot, rainOnehot.shape)
#需要修改：看ordinal网络的输出
#TODO: 应该已经改过了？但我觉得不应该用0.5，因为这个应该是间隔？
#但感觉regressionValue的具体数值并不重要，只要分类对了就行
#REPO: 我也觉得这个0.5可以不用乘
regressionValues = 0.5 * (torch.sum(
(predictValues > 0.5).float(), dim=1).view(-1, 1)) #(N, 1)
zeros = torch.zeros(regressionValues.size()).to(self.device)
#print('regression:', regressionValues, regressionValues.shape)
#print('zeros:', zeros, zeros.shape)
# 将confidence网络的预测进行mask，得到最终的输出
#TODO: 感觉这种算法不对啊，除非rainOnehot是(N, 1, 2)，否则得不到(N,1,1)，但我觉得是(N,2)
#REPO: 实验下来，rainOnehot是(N,2),后一项是（N,2,1），乘出来是(N,N,1)[matmul的特性，后一项多出的第一维会作为batch，剩余二维部分做矩阵乘法]
regressionValues = torch.matmul(
rainOnehot, #(N, 2)
torch.cat((zeros, regressionValues),
dim=1).unsqueeze(-1)).squeeze(-1)
#print('regression new:', regressionValues,regressionValues.shape)
#regressionValues = regressionValues.item()
#计算重建损失和预测与标签的损失
regressionLoss = nn.MSELoss()(regressionValues, rain)
reconstructLoss = nn.MSELoss()(reconstructValues, input)
#TODO: 这里要用上分类处理后的标签数据
#计算confidence损失
rainClassificationLoss = nn.CrossEntropyLoss()(rainPreds,
rainClass)
#print('losses: ', regressionLoss, reconstructLoss,rainClassificationLoss)
# TODO: 感觉这里可以改改，直接用序回归出的分类来算，而不用预测出的具体降水值;
# 如果觉得不太合理就改回去吧
# REPO: 但这样的话，confidence网络就没用了，我们最终还是要给test生成具体降水值
# predictNumpy = regressionValues.cpu().numpy()
# for i, threas in enumerate(tsthreas):
# tp[i] += np.sum(
# (gt_micaps >= threas) * (predictNumpy >= threas))
# tn[i] += np.sum(
# (gt_micaps < threas) * (predictNumpy < threas))
# fp[i] += np.sum(
# (gt_micaps < threas) * (predictNumpy >= threas))
# fn[i] += np.sum(
# (gt_micaps >= threas) * (predictNumpy < threas))
threshold_for_probability = 0.5
predictValues_numpy = predictValues.cpu().numpy()
#print(predictValues_numpy, gt_micaps)
#print(predictValues_numpy.shape)
for i, threas in enumerate(tsthreas):
tp[i] += np.sum(
(gt_micaps >= threas) * (predictValues_numpy[:, i] >=
threshold_for_probability))
tn[i] += np.sum(
(gt_micaps < threas) * (predictValues_numpy[:, i] <
threshold_for_probability))
fp[i] += np.sum(
(gt_micaps < threas) * (predictValues_numpy[:, i] >=
threshold_for_probability))
fn[i] += np.sum(
(gt_micaps >= threas) * (predictValues_numpy[:, i] <
threshold_for_probability))
#print('finals:', tp, tn, fp, fn)
#计算TS,四舍五入保留5位小数
for i, _ in enumerate(tsthreas):
ts[i] += round(tp[i] / (tp[i] + fp[i] + fn[i]), 5)
# save TS Scores to list[(_,_,_,_,_)]
tsDisplay = list(zip(tp, tn, fp, fn, ts))
"""---------------- print validation info ----------------"""
info = {
"ts_score": tsDisplay,
"test_rain_classification_accuracy": rainAccuracy,
}
print(
"========================== Epoch {} Test Result Show =========================="
.format(1))
print(info)
#生成将所有的-99999变成0,其他为1的mask
def get_mask(self, x):
zero = torch.zeros_like(x)
ones = torch.ones_like(x)
x = torch.where(x > -99999, ones, x)
x = torch.where(x == -99999, zero, x)
return x
def generateOneHot(self, softmax):
maxIdxs = torch.argmax(softmax, dim=1, keepdim=True).cpu().long()
oneHotMask = torch.zeros(softmax.shape, dtype=torch.float32)
oneHotMask = oneHotMask.scatter_(1, maxIdxs, 1.0)
#oneHotMask = oneHotMask.unsqueeze(-2)
return oneHotMask
if __name__ == '__main__':
config = yaml.load(open('config.yaml', 'r'), Loader=yaml.FullLoader)
evaluate_dataset = gridNewDataset(config['train_dir'],
isTrain=False,
isFirstTime=False)
evaluate_iter = DataLoader(evaluate_dataset,
batch_size=256,
shuffle=True,
pin_memory=True)
device = 'cuda'
validate = Validate(config['combinatotorial'], evaluate_iter,
device).to(device)
validate.initialize('checkpoint/confidence2.pth',
'checkpoint/encoderwithodr3.pth',
'checkpoint/decoder.pth', 'checkpoint/odr3.pth')
#validate.forward()
validate.simple_validate()