-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathdataset.py
More file actions
748 lines (595 loc) · 31 KB
/
Copy pathdataset.py
File metadata and controls
748 lines (595 loc) · 31 KB
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
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
#%%
import torch
from torch.utils.data import Dataset
import numpy as np
import xarray as xr
import pickle
import pandas as pd
#%%
class UNetDataset(Dataset):
def __init__(self, cfg, fvarLst, pvarLst, ovarLst, fileidx, tRange, is_train=True):
super(UNetDataset, self).__init__()
self.dataPath = cfg.input_path
#self.df = xr.open_dataset(cfg.input_path + 'forcings_short.nc')
self.df = xr.open_dataset(cfg.input_path + 'forcings_day.nc')
self.dp = xr.open_dataset(cfg.input_path + 'par2000.nc')
self.do = xr.open_dataset(cfg.input_path + 'sm2000.nc')
#self.do = xr.open_dataset(cfg.input_path + 'e2000.nc')
self.nx = len(fvarLst) + len(pvarLst) + 1 # 输入维度：强迫变量数 + 参数变量数 + 1（初始状态变量）
self.ny = len(ovarLst) # 输出维度：目标变量数（这里是蒸散发变量数）(现在改成sm一层的了)
self.nfile = fileidx[1]-fileidx[0]
self.fileidx = fileidx #文件索引范围
self.tRange = tRange
self.nt = len(self.do.time.loc[tRange[0]:tRange[1]])
self.is_train = is_train
if self.is_train:
self.stat = {}
else:
scaler_file = cfg.out_dir + "train_data_scaler.bin"
with open(scaler_file, mode='rb') as fp:
self.stat = pickle.load(fp)
# （注释代码）备用：加载预训练的统计量（如AE模型的统计量，用于迁移学习）
# with open('/home/sunrc/work/VICcases/huaihe_new/surrogatedL/HPCresults/ARnet/AE_7pnewin/train_data_scaler.bin', mode='rb') as fp:
# self.statini = pickle.load(fp)
self.forcings = self.getDataTs(self.df, fvarLst)
self.pars = self.getDataConst(self.dp, pvarLst)
self.target = self.getDataTs(self.do, ovarLst) #后面有定义这三种处理函数
#初始状态变量
moist_ini = np.full((self.nfile, 1, 1, len(self.do.lat), len(self.do.lon)), 20)
moist_ini = (moist_ini -self.stat['OUT_SOIL_MOIST_mean'])/self.stat['OUT_SOIL_MOIST_std']
#moist_ini = np.full((self.nfile, 1, 1, len(self.do.lat), len(self.do.lon)), 0.5)
#moist_ini = (moist_ini -self.stat['OUT_EVAP_mean'])/self.stat['OUT_EVAP_std'] #归一化
self.target= np.concatenate([moist_ini, self.target], axis=1) #拼接目标变量：将初始状态（1个时间步）拼接到目标变量的时间维度前
if self.is_train: # 训练集：保存统计量到文件
file_path = cfg.out_dir + "train_data_scaler.bin"
with open(file_path, mode='wb') as fp:
pickle.dump(self.stat, fp)
lookup = [(i,k) for i in range(self.nfile) for k in range(self.nt)]
self.lookup_table = {i: elem for i, elem in enumerate(lookup)} #构建样本索引表
def __len__(self):
return len(self.lookup_table)
def __getitem__(self, idx):
file, indices = self.lookup_table[idx] # 从索引表获取（文件i，时间步k）
input = np.concatenate([self.forcings[0, indices, :, :, :], self.pars[file, indices, :, :, :], self.target[file, indices, :, :, :] ], axis=0) #targets的 indices 是上一时刻，因为 self.target 包含了初始状态
output = self.target[file, indices+1, :, :, :]
input = torch.from_numpy(input).float()
output = torch.from_numpy(output).float()
return input, output
def getDataTs(self, ds, varLst):
if 'nfiles' not in ds.dims: #维度名判断文件数
nfile = 1
else:
nfile = self.nfile
nvar = len(varLst)
nt = self.nt
data = np.ndarray([nfile, nt, nvar, len(ds.lat), len(ds.lon)])
for k in range(nvar):
var = ds[varLst[k]]
if len(var.dims) > 4: # model output(soil moisture)
dataTemp = var.loc[self.fileidx[0]:self.fileidx[1],self.tRange[0]:self.tRange[1],0,:,:].values ## only use the first layer of output data
#if len(var.dims) > 3: # model output(ET)
# dataTemp = var.loc[self.fileidx[0]:self.fileidx[1],self.tRange[0]:self.tRange[1],:,:].values
else:
dataTemp = var.loc[self.tRange[0]:self.tRange[1],:,:].values
if self.is_train:
mean = np.nanmean(dataTemp)
std = np.nanstd(dataTemp)
self.stat[varLst[k]+'_mean'] = mean
self.stat[varLst[k]+'_std'] = std
else:
mean = self.stat[varLst[k]+'_mean']
std = self.stat[varLst[k]+'_std']
dataTemp = (dataTemp-mean)/std
if len(var.dims) == 3: # forcing
data[0, :, k, :, :] = dataTemp
else:
data[:, :, k, :, :] = dataTemp
return data
def getDataConst(self, ds, varLst):
nvar = len(varLst)
nfile = self.nfile
nt = self.nt
data = np.ndarray([nfile, nvar, len(ds.lat), len(ds.lon)])
for k in range(nvar):
if varLst[k] == 'expt':
dataTemp = ds[varLst[k]].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],1,:,:] # second layer
elif varLst[k] == 'd2':
dataTemp = ds['depth'].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],1,:,:] # second layer
elif varLst[k] == 'd3':
dataTemp = ds['depth'].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],2,:,:] # third layer
else:
dataTemp = ds[varLst[k]].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],:,:]
if self.is_train:
mean = np.nanmean(dataTemp)
std = np.nanstd(dataTemp)
self.stat[varLst[k]+'_mean'] = mean
self.stat[varLst[k]+'_std'] = std
else:
mean = self.stat[varLst[k]+'_mean']
std = self.stat[varLst[k]+'_std']
dataTemp = (dataTemp-mean)/std
# dataTemp = (dataTemp - self.statini[varLst[k]+'_mean'])/self.statini[varLst[k]+'_std']
data[:, k, :, :] = dataTemp
out = np.repeat(np.reshape(data, [nfile, 1, nvar, len(ds.lat), len(ds.lon)]), nt, axis=1)
return out
#%%
class ResNetDataset(Dataset):
def __init__(self, cfg, fvarLst, pvarLst, ovarLst, fileidx, tRange, is_train=True):
super(ResNetDataset, self).__init__()
self.dataPath = cfg.input_path
#self.df = xr.open_dataset(cfg.input_path + 'forcings_short.nc')
self.df = xr.open_dataset(cfg.input_path + 'forcings_day.nc')
self.dp = xr.open_dataset(cfg.input_path + 'par2000.nc')
self.do = xr.open_dataset(cfg.input_path + 'sm2000.nc')
#self.do = xr.open_dataset(cfg.input_path + 'e2000.nc')
self.nx = len(fvarLst) + len(pvarLst) + 1 # 输入维度：强迫变量数 + 参数变量数 + 1（初始状态变量）
self.ny = len(ovarLst) # 输出维度：目标变量数（这里是蒸散发变量数）(现在改成sm一层的了)
self.nfile = fileidx[1]-fileidx[0]
self.fileidx = fileidx #文件索引范围
self.tRange = tRange
self.nt = len(self.do.time.loc[tRange[0]:tRange[1]])
self.is_train = is_train
if self.is_train:
self.stat = {}
else:
scaler_file = cfg.out_dir + "train_data_scaler.bin"
with open(scaler_file, mode='rb') as fp:
self.stat = pickle.load(fp)
# （注释代码）备用：加载预训练的统计量（如AE模型的统计量，用于迁移学习）
# with open('/home/sunrc/work/VICcases/huaihe_new/surrogatedL/HPCresults/ARnet/AE_7pnewin/train_data_scaler.bin', mode='rb') as fp:
# self.statini = pickle.load(fp)
self.forcings = self.getDataTs(self.df, fvarLst)
self.pars = self.getDataConst(self.dp, pvarLst)
self.target = self.getDataTs(self.do, ovarLst) #后面有定义这三种处理函数
#初始状态变量
moist_ini = np.full((self.nfile, 1, 1, len(self.do.lat), len(self.do.lon)), 20)
moist_ini = (moist_ini -self.stat['OUT_SOIL_MOIST_mean'])/self.stat['OUT_SOIL_MOIST_std']
#moist_ini = np.full((self.nfile, 1, 1, len(self.do.lat), len(self.do.lon)), 0.5)
#moist_ini = (moist_ini -self.stat['OUT_EVAP_mean'])/self.stat['OUT_EVAP_std'] #归一化
self.target= np.concatenate([moist_ini, self.target], axis=1) #拼接目标变量：将初始状态（1个时间步）拼接到目标变量的时间维度前
if self.is_train: # 训练集：保存统计量到文件
file_path = cfg.out_dir + "train_data_scaler.bin"
with open(file_path, mode='wb') as fp:
pickle.dump(self.stat, fp)
lookup = [(i,k) for i in range(self.nfile) for k in range(self.nt)]
self.lookup_table = {i: elem for i, elem in enumerate(lookup)} #构建样本索引表
def __len__(self):
return len(self.lookup_table)
def __getitem__(self, idx):
file, indices = self.lookup_table[idx] # 从索引表获取（文件i，时间步k）
input = np.concatenate([self.forcings[0, indices, :, :, :], self.pars[file, indices, :, :, :], self.target[file, indices, :, :, :] ], axis=0) #targets的 indices 是上一时刻，因为 self.target 包含了初始状态
output = self.target[file, indices+1, :, :, :]
input = torch.from_numpy(input).float()
output = torch.from_numpy(output).float()
return input, output
def getDataTs(self, ds, varLst):
if 'nfiles' not in ds.dims: #维度名判断文件数
nfile = 1
else:
nfile = self.nfile
nvar = len(varLst)
nt = self.nt
data = np.ndarray([nfile, nt, nvar, len(ds.lat), len(ds.lon)])
for k in range(nvar):
var = ds[varLst[k]]
if len(var.dims) > 4: # model output(soil moisture)
dataTemp = var.loc[self.fileidx[0]:self.fileidx[1],self.tRange[0]:self.tRange[1],0,:,:].values ## only use the first layer of output data
#if len(var.dims) > 3: # model output(ET)
# dataTemp = var.loc[self.fileidx[0]:self.fileidx[1],self.tRange[0]:self.tRange[1],:,:].values
else:
dataTemp = var.loc[self.tRange[0]:self.tRange[1],:,:].values
if self.is_train:
mean = np.nanmean(dataTemp)
std = np.nanstd(dataTemp)
self.stat[varLst[k]+'_mean'] = mean
self.stat[varLst[k]+'_std'] = std
else:
mean = self.stat[varLst[k]+'_mean']
std = self.stat[varLst[k]+'_std']
dataTemp = (dataTemp-mean)/std
if len(var.dims) == 3: # forcing
data[0, :, k, :, :] = dataTemp
else:
data[:, :, k, :, :] = dataTemp
return data
def getDataConst(self, ds, varLst):
nvar = len(varLst)
nfile = self.nfile
nt = self.nt
data = np.ndarray([nfile, nvar, len(ds.lat), len(ds.lon)])
for k in range(nvar):
if varLst[k] == 'expt':
dataTemp = ds[varLst[k]].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],1,:,:] # second layer
elif varLst[k] == 'd2':
dataTemp = ds['depth'].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],1,:,:] # second layer
elif varLst[k] == 'd3':
dataTemp = ds['depth'].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],2,:,:] # third layer
else:
dataTemp = ds[varLst[k]].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],:,:]
if self.is_train:
mean = np.nanmean(dataTemp)
std = np.nanstd(dataTemp)
self.stat[varLst[k]+'_mean'] = mean
self.stat[varLst[k]+'_std'] = std
else:
mean = self.stat[varLst[k]+'_mean']
std = self.stat[varLst[k]+'_std']
dataTemp = (dataTemp-mean)/std
# dataTemp = (dataTemp - self.statini[varLst[k]+'_mean'])/self.statini[varLst[k]+'_std']
data[:, k, :, :] = dataTemp
out = np.repeat(np.reshape(data, [nfile, 1, nvar, len(ds.lat), len(ds.lon)]), nt, axis=1)
return out
# %%
class ARDenseNetDataset(Dataset):
def __init__(self, cfg, fvarLst, pvarLst, ovarLst, fileidx, tRange, is_train=True):
super(ARDenseNetDataset, self).__init__()
self.dataPath = cfg.input_path
#self.df = xr.open_dataset(cfg.input_path + 'forcings_short.nc')
self.df = xr.open_dataset(cfg.input_path + 'forcings_day.nc')
self.dp = xr.open_dataset(cfg.input_path + 'par2000.nc')
self.do = xr.open_dataset(cfg.input_path + 'sm2000.nc')
#self.do = xr.open_dataset(cfg.input_path + 'e2000.nc')
self.nx = len(fvarLst) + len(pvarLst) + 1 # 输入维度：强迫变量数 + 参数变量数 + 1（初始状态变量）
self.ny = len(ovarLst) # 输出维度：目标变量数（这里是蒸散发变量数）(现在改成sm一层的了)
self.nfile = fileidx[1]-fileidx[0]
self.fileidx = fileidx #文件索引范围
self.tRange = tRange
self.nt = len(self.do.time.loc[tRange[0]:tRange[1]])
self.is_train = is_train
if self.is_train:
self.stat = {}
else:
scaler_file = cfg.out_dir + "train_data_scaler.bin"
with open(scaler_file, mode='rb') as fp:
self.stat = pickle.load(fp)
# （注释代码）备用：加载预训练的统计量（如AE模型的统计量，用于迁移学习）
# with open('/home/sunrc/work/VICcases/huaihe_new/surrogatedL/HPCresults/ARnet/AE_7pnewin/train_data_scaler.bin', mode='rb') as fp:
# self.statini = pickle.load(fp)
self.forcings = self.getDataTs(self.df, fvarLst)
self.pars = self.getDataConst(self.dp, pvarLst)
self.target = self.getDataTs(self.do, ovarLst) #后面有定义这三种处理函数
#初始状态变量
moist_ini = np.full((self.nfile, 1, 1, len(self.do.lat), len(self.do.lon)), 20)
moist_ini = (moist_ini -self.stat['OUT_SOIL_MOIST_mean'])/self.stat['OUT_SOIL_MOIST_std']
#moist_ini = np.full((self.nfile, 1, 1, len(self.do.lat), len(self.do.lon)), 0.5)
#moist_ini = (moist_ini -self.stat['OUT_EVAP_mean'])/self.stat['OUT_EVAP_std'] #归一化
self.target= np.concatenate([moist_ini, self.target], axis=1) #拼接目标变量：将初始状态（1个时间步）拼接到目标变量的时间维度前
if self.is_train: # 训练集：保存统计量到文件
file_path = cfg.out_dir + "train_data_scaler.bin"
with open(file_path, mode='wb') as fp:
pickle.dump(self.stat, fp)
lookup = [(i,k) for i in range(self.nfile) for k in range(self.nt)]
self.lookup_table = {i: elem for i, elem in enumerate(lookup)} #构建样本索引表
def __len__(self):
return len(self.lookup_table)
def __getitem__(self, idx):
file, indices = self.lookup_table[idx] # 从索引表获取（文件i，时间步k）
input = np.concatenate([self.forcings[0, indices, :, :, :], self.pars[file, indices, :, :, :], self.target[file, indices, :, :, :] ], axis=0) #targets的 indices 是上一时刻，因为 self.target 包含了初始状态
output = self.target[file, indices+1, :, :, :]
input = torch.from_numpy(input).float()
output = torch.from_numpy(output).float()
return input, output
def getDataTs(self, ds, varLst):
if 'nfiles' not in ds.dims: #维度名判断文件数
nfile = 1
else:
nfile = self.nfile
nvar = len(varLst)
nt = self.nt
data = np.ndarray([nfile, nt, nvar, len(ds.lat), len(ds.lon)])
for k in range(nvar):
var = ds[varLst[k]]
if len(var.dims) > 4: # model output(soil moisture)
dataTemp = var.loc[self.fileidx[0]:self.fileidx[1],self.tRange[0]:self.tRange[1],0,:,:].values ## only use the first layer of output data
#if len(var.dims) > 3: # model output(ET)
# dataTemp = var.loc[self.fileidx[0]:self.fileidx[1],self.tRange[0]:self.tRange[1],:,:].values
else:
dataTemp = var.loc[self.tRange[0]:self.tRange[1],:,:].values
if self.is_train:
mean = np.nanmean(dataTemp)
std = np.nanstd(dataTemp)
self.stat[varLst[k]+'_mean'] = mean
self.stat[varLst[k]+'_std'] = std
else:
mean = self.stat[varLst[k]+'_mean']
std = self.stat[varLst[k]+'_std']
dataTemp = (dataTemp-mean)/std
if len(var.dims) == 3: # forcing
data[0, :, k, :, :] = dataTemp
else:
data[:, :, k, :, :] = dataTemp
return data
def getDataConst(self, ds, varLst):
nvar = len(varLst)
nfile = self.nfile
nt = self.nt
data = np.ndarray([nfile, nvar, len(ds.lat), len(ds.lon)])
for k in range(nvar):
if varLst[k] == 'expt':
dataTemp = ds[varLst[k]].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],1,:,:] # second layer
elif varLst[k] == 'd2':
dataTemp = ds['depth'].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],1,:,:] # second layer
elif varLst[k] == 'd3':
dataTemp = ds['depth'].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],2,:,:] # third layer
else:
dataTemp = ds[varLst[k]].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1],:,:]
if self.is_train:
mean = np.nanmean(dataTemp)
std = np.nanstd(dataTemp)
self.stat[varLst[k]+'_mean'] = mean
self.stat[varLst[k]+'_std'] = std
else:
mean = self.stat[varLst[k]+'_mean']
std = self.stat[varLst[k]+'_std']
dataTemp = (dataTemp-mean)/std
# dataTemp = (dataTemp - self.statini[varLst[k]+'_mean'])/self.statini[varLst[k]+'_std']
data[:, k, :, :] = dataTemp
out = np.repeat(np.reshape(data, [nfile, 1, nvar, len(ds.lat), len(ds.lon)]), nt, axis=1)
return out
#%%
#class FSTRDataset(Dataset):
# def __init__(self, cfg, fvarLst, pvarLst, ovarLst, fileidx, tRange, is_train=True):
# super(FSTRDataset, self).__init__()
# self.dataPath = cfg.input_path
# self.df = xr.open_dataset(cfg.input_path + 'forcings_day.nc')
# self.dp = xr.open_dataset(cfg.input_path + 'par2000.nc')
# self.do = xr.open_dataset(cfg.input_path + 'sm2000.nc')
#
# self.nx = len(fvarLst) + len(pvarLst) + 1
# self.ny = len(ovarLst)
# self.nfile = fileidx[1] - fileidx[0]
# self.fileidx = fileidx
# self.tRange = tRange
# self.nt = len(self.do.time.loc[tRange[0]:tRange[1]])
# self.is_train = is_train
#
# if self.is_train:
# self.stat = {}
# else:
# scaler_file = cfg.out_dir + "train_data_scaler.bin"
# with open(scaler_file, mode='rb') as fp:
# self.stat = pickle.load(fp)
#
# 获取数据
# self.forcings = self.getDataTs(self.df, fvarLst)
# self.pars = self.getDataConst(self.dp, pvarLst)
# self.target = self.getDataTs(self.do, ovarLst)
# # 初始状态变量
# moist_ini = np.full((self.nfile, 1, 1, len(self.do.lat), len(self.do.lon)), 20)
# moist_ini = (moist_ini - self.stat.get('OUT_SOIL_MOIST_mean', 0)) / self.stat.get('OUT_SOIL_MOIST_std', 1)
#
# self.target = np.concatenate([moist_ini, self.target], axis=1)
#
# if self.is_train:
# file_path = cfg.out_dir + "train_data_scaler.bin"
# with open(file_path, mode='wb') as fp:
# pickle.dump(self.stat, fp)
#
# 创建样本索引
# lookup = [(i, k) for i in range(self.nfile) for k in range(self.nt)]
# self.lookup_table = {i: elem for i, elem in enumerate(lookup)}
# def __len__(self):
# return len(self.lookup_table)
# def __getitem__(self, idx):
# """返回三个输入和一个输出
# Returns:
# init_cond: (1, H, W) - 上一时刻的土壤湿度
# forcings: (7, H, W) - 当前时刻的强迫变量
# static_inputs: (7, H, W) - 静态参数
# output: (1, H, W) - 当前时刻的土壤湿度
# """
# file, indices = self.lookup_table[idx]
#
# # 输入1: 上一时刻的土壤湿度 (init_cond)
# init_cond = self.target[file, indices, :, :, :]
# # 输入2: 当前时刻的强迫变量 (forcings)
# forcings = self.forcings[0, indices, :, :, :]
# # 输入3: 静态参数 (static_inputs)
# static_inputs = self.pars[file, indices, :, :, :]
#
# 输出: 当前时刻的土壤湿度
# output = self.target[file, indices + 1, :, :, :]
# 转换为torch张量
# init_cond = torch.from_numpy(init_cond).float()
# forcings = torch.from_numpy(forcings).float()
# static_inputs = torch.from_numpy(static_inputs).float()
# output = torch.from_numpy(output).float()
# return init_cond, forcings, static_inputs, output
# def getDataTs(self, ds, varLst):
# """获取时间序列数据"""
# if 'nfiles' not in ds.dims:
# nfile = 1
# else:
# nfile = self.nfile
# nvar = len(varLst)
# nt = self.nt
# data = np.ndarray([nfile, nt, nvar, len(ds.lat), len(ds.lon)])
# for k in range(nvar):
# var = ds[varLst[k]]
# if len(var.dims) > 4: # model output (soil moisture)
# dataTemp = var.loc[self.fileidx[0]:self.fileidx[1], self.tRange[0]:self.tRange[1], 0, :, :].values
# else:
# dataTemp = var.loc[self.tRange[0]:self.tRange[1], :, :].values
# if self.is_train:
# mean = np.nanmean(dataTemp)
# std = np.nanstd(dataTemp)
# self.stat[varLst[k] + '_mean'] = mean
# self.stat[varLst[k] + '_std'] = std
# else:
# mean = self.stat[varLst[k] + '_mean']
# std = self.stat[varLst[k] + '_std']
# dataTemp = (dataTemp - mean) / std
# if len(var.dims) == 3: # forcing
# data[0, :, k, :, :] = dataTemp
# else:
# data[:, :, k, :, :] = dataTemp
# return data
# def getDataConst(self, ds, varLst):
# """获取静态参数数据"""
# nvar = len(varLst)
# nfile = self.nfile
# nt = self.nt
# data = np.ndarray([nfile, nvar, len(ds.lat), len(ds.lon)])
# for k in range(nvar):
# if varLst[k] == 'expt':
# dataTemp = ds[varLst[k]].values
# dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1], 1, :, :]
# elif varLst[k] == 'd2':
# dataTemp = ds['depth'].values
# dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1], 1, :, :]
# elif varLst[k] == 'd3':
# dataTemp = ds['depth'].values
# dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1], 2, :, :]
# else:
# dataTemp = ds[varLst[k]].values
# dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1], :, :]
# if self.is_train:
# mean = np.nanmean(dataTemp)
# std = np.nanstd(dataTemp)
# self.stat[varLst[k] + '_mean'] = mean
# self.stat[varLst[k] + '_std'] = std
# else:
# mean = self.stat[varLst[k] + '_mean']
# std = self.stat[varLst[k] + '_std']
# dataTemp = (dataTemp - mean) / std
# data[:, k, :, :] = dataTemp
# 扩展到时间维度
# out = np.repeat(np.reshape(data, [nfile, 1, nvar, len(ds.lat), len(ds.lon)]), nt, axis=1)
# return out
#%%
#%%
class FSTRDataset(Dataset):
def __init__(self, cfg, fvarLst, pvarLst, ovarLst, fileidx, tRange, seq_len, is_train=True):
super(FSTRDataset, self).__init__()
self.fvarLst = fvarLst # 动态变量名列表
self.pvarLst = pvarLst # 静态参数名列表
self.ovarLst = ovarLst # 目标变量名列表
self.dataPath = cfg.input_path
self.df = xr.open_dataset(cfg.input_path + 'forcings_day.nc')
self.dp = xr.open_dataset(cfg.input_path + 'par2000.nc')
self.do = xr.open_dataset(cfg.input_path + 'sm2000.nc')
self.nfile = fileidx[1] - fileidx[0]
self.fileidx = fileidx
self.tRange = tRange
self.is_train = is_train
# 计算原始时间范围内的天数（要预测的天数）
start = pd.Timestamp(tRange[0])
end = pd.Timestamp(tRange[1])
self.nt = len(pd.date_range(start, end))
self.seq_len = seq_len
# target 需要包含前一天
init_date = (start - pd.Timedelta(days=1)).strftime('%Y-%m-%d')
target_tRange = [init_date, tRange[1]] # 长度 = nt + 1
if self.is_train:
self.stat = {}
else:
scaler_file = cfg.out_dir + "train_data_scaler.bin"
with open(scaler_file, mode='rb') as fp:
self.stat = pickle.load(fp)
# 读取数据：forcings 和 pars 只用原始 tRange
self.forcings = self.getDataTs(self.df, fvarLst, time_range=tRange) # [nfile, nt, fvar_len, H, W]
self.pars = self.getDataConst(self.dp, pvarLst, time_range=tRange) # [nfile, nt, pvar_len, H, W]
# target 使用扩展范围
self.target = self.getDataTs(self.do, ovarLst, time_range=target_tRange) # [nfile, nt+1, out_len, H, W]
if self.is_train:
file_path = cfg.out_dir + "train_data_scaler.bin"
with open(file_path, mode='wb') as fp:
pickle.dump(self.stat, fp)
# 构建样本索引：offset 为起始预测日相对于 start 的偏移，取值范围 0 到 nt - seq_len
lookup = [(i, offset) for i in range(self.nfile) for offset in range(0, self.nt - self.seq_len + 1)]
self.lookup_table = {i: elem for i, elem in enumerate(lookup)}
def __len__(self):
return len(self.lookup_table)
def __getitem__(self, idx):
file, offset = self.lookup_table[idx] # offset: 起始预测日相对于 start 的偏移
# 前一时刻土壤湿度（初始条件）：第 start+offset-1 天的真实值，对应 target 索引 offset
prev_sm = self.target[file, offset] # shape: (out_len, H, W)
# 连续 seq_len 个时刻的气象强迫：从 start+offset 到 start+offset+seq_len-1
forcings = self.forcings[file, offset : offset+self.seq_len] # shape: (seq_len, fvar_len, H, W)
# 静态参数（任意时刻，取 offset 时刻）
static_inputs = self.pars[file, offset] # shape: (pvar_len, H, W)
# 连续 seq_len 个目标值：从 start+offset 到 start+offset+seq_len-1
target = self.target[file, offset+1 : offset+self.seq_len+1] # shape: (seq_len, out_len, H, W)
# 转换为 tensor
prev_sm = torch.from_numpy(prev_sm).float().unsqueeze(0) # (1, out_len, H, W)
forcings = torch.from_numpy(forcings).float() # (seq_len, fvar_len, H, W)
static_inputs = torch.from_numpy(static_inputs).float().unsqueeze(0) # (1, pvar_len, H, W)
target = torch.from_numpy(target).float() # (seq_len, out_len, H, W)
return prev_sm, forcings, static_inputs, target
def getDataTs(self, ds, varLst, time_range):
"""获取时间序列数据，返回 [nfile, nt, nvar, lat, lon]"""
if len(ds.dims) == 3: # 强迫：只有时间维
nfile = 1
else:
nfile = self.nfile
nvar = len(varLst)
# 计算时间范围内的天数
nt_total = len(pd.date_range(time_range[0], time_range[1]))
data = np.ndarray([nfile, nt_total, nvar, len(ds.lat), len(ds.lon)])
for k in range(nvar):
var = ds[varLst[k]]
if len(ds.dims) > 4: # model output [file, time, layer, lat, lon]
dataTemp = var.loc[self.fileidx[0]:self.fileidx[1], time_range[0]:time_range[1], 0, :, :].values
else:
dataTemp = var.loc[time_range[0]:time_range[1], :, :].values
if self.is_train:
mean = np.nanmean(dataTemp)
std = np.nanstd(dataTemp)
self.stat[varLst[k] + '_mean'] = mean
self.stat[varLst[k] + '_std'] = std
else:
mean = self.stat[varLst[k] + '_mean']
std = self.stat[varLst[k] + '_std']
dataTemp = (dataTemp - mean) / std
if len(ds.dims) == 4: # forcing
# 对于 forcing，所有文件共用第0个文件的数据
data[0, :, k, :, :] = dataTemp
else:
data[:, :, k, :, :] = dataTemp
return data
def getDataConst(self, ds, varLst, time_range):
"""获取静态参数数据，返回 [nfile, nt, nvar, lat, lon]（时间维重复）"""
nvar = len(varLst)
nfile = self.nfile
nt_total = len(pd.date_range(time_range[0], time_range[1]))
# 先读取静态参数，形状 [nfile, nvar, lat, lon]
data_static = np.ndarray([nfile, nvar, len(ds.lat), len(ds.lon)])
for k in range(nvar):
if varLst[k] == 'expt':
dataTemp = ds[varLst[k]].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1], 1, :, :]
elif varLst[k] == 'd2':
dataTemp = ds['depth'].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1], 1, :, :]
elif varLst[k] == 'd3':
dataTemp = ds['depth'].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1], 2, :, :]
else:
dataTemp = ds[varLst[k]].values
dataTemp = dataTemp[self.fileidx[0]:self.fileidx[1], :, :]
if self.is_train:
mean = np.nanmean(dataTemp)
std = np.nanstd(dataTemp)
self.stat[varLst[k] + '_mean'] = mean
self.stat[varLst[k] + '_std'] = std
else:
mean = self.stat[varLst[k] + '_mean']
std = self.stat[varLst[k] + '_std']
dataTemp = (dataTemp - mean) / std
data_static[:, k, :, :] = dataTemp
# 在时间维上重复到 nt_total
out = np.repeat(np.reshape(data_static, [nfile, 1, nvar, len(ds.lat), len(ds.lon)]), nt_total, axis=1)
return out
# %%