-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdatasets.py
515 lines (448 loc) · 19.8 KB
/
datasets.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
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
'''
Copied from https://github.com/JiaruiFeng/KP-GNN/tree/a127847ed8aa2955f758476225bc27c6697e7733
'''
from sklearn.metrics import accuracy_score
import torch
import pickle
import numpy as np
import scipy.io as sio
from scipy.special import comb
import networkx as nx
import numpy as np
from torch_geometric.data import InMemoryDataset, Data
from torch_geometric.utils import to_undirected, degree
from torch_geometric.datasets import TUDataset, ZINC, GNNBenchmarkDataset, QM9
from ogb.graphproppred import PygGraphPropPredDataset, Evaluator
from data_I2GNN import dataset_random_graph
class PlanarSATPairsDataset(InMemoryDataset):
def __init__(self,
root="dataset/EXP",
transform=None,
pre_transform=None,
pre_filter=None):
super(PlanarSATPairsDataset, self).__init__(root, transform,
pre_transform, pre_filter)
data, slices = torch.load(self.processed_paths[0])
if data.x.dim() == 1:
data.x = data.x.unsqueeze(-1)
data.x += 1
data.y = data.y.to(torch.float).reshape(-1, 1)
self.data, self.slices = data, slices
@property
def raw_file_names(self):
return ["GRAPHSAT" + ".pkl"]
@property
def processed_file_names(self):
return 'data.pt'
def download(self):
pass
def process(self):
# Read data into huge `Data` list.
with open("dataset/EXP/raw/" + "newGRAPHSAT" + ".pkl", "rb") as f:
data_list = pickle.load(f)
data_list = [Data.from_dict(_) for _ in data_list]
if self.pre_filter is not None:
data_list = [data for data in data_list if self.pre_filter(data)]
if self.pre_transform is not None:
data_list = [self.pre_transform(data) for data in data_list]
data, slices = self.collate(data_list)
torch.save((data, slices), self.processed_paths[0])
def EXP_node_feature_transform(data):
data.x = data.x[:, 0].to(torch.long)
return data
class GraphCountDataset(InMemoryDataset):
def __init__(self,
root="dataset/subgraphcount",
transform=None,
pre_transform=None):
super(GraphCountDataset, self).__init__(root, transform, pre_transform)
self.data, self.slices = torch.load(self.processed_paths[0])
a = sio.loadmat(self.raw_paths[0])
self.train_idx = torch.from_numpy(a['train_idx'][0])
self.val_idx = torch.from_numpy(a['val_idx'][0])
self.test_idx = torch.from_numpy(a['test_idx'][0])
@property
def raw_file_names(self):
return ["randomgraph.mat"]
@property
def processed_file_names(self):
return 'data.pt'
def download(self):
# Download to `self.raw_dir`.
pass
def process(self):
# Read data into huge `Data` list.
b = self.processed_paths[0]
a = sio.loadmat(self.raw_paths[0]) # 'subgraphcount/randomgraph.mat')
# list of adjacency matrix
A = a['A'][0]
# list of output
Y = a['F']
data_list = []
for i in range(len(A)):
a = A[i]
A2 = a.dot(a)
A3 = A2.dot(a)
tri = np.trace(A3) / 6
tailed = ((np.diag(A3) / 2) * (a.sum(0) - 2)).sum()
cyc4 = 1 / 8 * (np.trace(A3.dot(a)) + np.trace(A2) - 2 * A2.sum())
cus = a.dot(np.diag(np.exp(-a.dot(a).sum(1)))).dot(a).sum()
deg = a.sum(0)
star = 0
for j in range(a.shape[0]):
star += comb(int(deg[j]), 3)
expy = torch.tensor([[tri, tailed, star, cyc4, cus]])
E = np.where(A[i] > 0)
edge_index = torch.Tensor(np.vstack(
(E[0], E[1]))).type(torch.int64)
x = torch.ones(A[i].shape[0], 1).long() # change to category
# y=torch.tensor(Y[i:i+1,:])
data_list.append(Data(edge_index=edge_index, x=x, y=expy))
if self.pre_filter is not None:
data_list = [data for data in data_list if self.pre_filter(data)]
if self.pre_transform is not None:
data_list = [self.pre_transform(data) for data in data_list]
data, slices = self.collate(data_list)
torch.save((data, slices), self.processed_paths[0])
class SRDataset(InMemoryDataset):
def __init__(self,
root="dataset/sr25",
transform=None,
pre_transform=None):
super(SRDataset, self).__init__(root, transform, pre_transform)
data, slices = torch.load(self.processed_paths[0])
data.x = data.x.long() + 1
data.y = torch.arange(data.y.shape[0])
self.data, self.slices = data, slices
@property
def raw_file_names(self):
return ["sr251256.g6"] # sr251256 sr351668
@property
def processed_file_names(self):
return 'data.pt'
def download(self):
# Download to `self.raw_dir`.
pass
def process(self):
# Read data into huge `Data` list.
dataset = nx.read_graph6(self.raw_paths[0])
data_list = []
for i, datum in enumerate(dataset):
x = torch.ones(datum.number_of_nodes(), 1)
edge_index = to_undirected(
torch.tensor(list(datum.edges())).transpose(1, 0))
data_list.append(Data(edge_index=edge_index, x=x, y=0))
if self.pre_filter is not None:
data_list = [data for data in data_list if self.pre_filter(data)]
if self.pre_transform is not None:
data_list = [self.pre_transform(data) for data in data_list]
data, slices = self.collate(data_list)
torch.save((data, slices), self.processed_paths[0])
class myEvaluator(Evaluator):
def __init__(self, name):
super().__init__(name=name)
def __call__(self, y_pred, y_true):
ret = super().eval({"y_pred": y_pred, "y_true": y_true})
assert len(ret) == 1
return list(ret.values())[0]
from torchmetrics import Accuracy, MeanAbsoluteError, F1Score
from torchmetrics.classification import MultilabelAveragePrecision
from typing import Iterable, Callable, Optional, Tuple
from torch_geometric.data import Dataset
from torch_geometric.datasets import LRGBDataset
def loaddataset(name: str,
**kwargs): #-> Iterable[Dataset], str, Callable, str
if name == "sr":
dataset = SRDataset(**kwargs)
dataset.num_tasks = torch.max(dataset.data.y).item() + 1
return (dataset, dataset, dataset), "fixed", Accuracy(
"multiclass",
num_classes=dataset.num_tasks), "cls" # full training/valid/test??
elif name == "EXP":
dataset = PlanarSATPairsDataset(
pre_transform=EXP_node_feature_transform, **kwargs)
dataset.num_tasks = 1
return (dataset, ), "fold-8-1-1", Accuracy("binary"), "bincls"
elif name == "CSL":
def CSL_node_feature_transform(data):
if "x" not in data:
data.x = torch.ones([data.num_nodes, 1], dtype=torch.float)
return data
dataset = GNNBenchmarkDataset("dataset",
"CSL",
pre_transform=CSL_node_feature_transform,
**kwargs)
dataset.num_tasks = torch.max(dataset.y).item() + 1
return (dataset, ), "fold-8-1-1", Accuracy("multiclass",
num_classes=10), "cls"
elif name.startswith("subgcount"):
y_slice = int(name[len("subgcount"):])
dataset = GraphCountDataset(**kwargs)
dataset.data.y = dataset.data.y - dataset.data.y.mean(dim=0)
dataset.data.y = dataset.data.y / dataset.data.y.std(dim=0)
dataset.data.y = dataset.data.y[:, [y_slice]]
# degree feature
# dataset.data.x.copy_(torch.cat([degree(dat.edge_index[0], num_nodes=dat.num_nodes, dtype=torch.long) for dat in dataset]).reshape(-1, 1))
dataset.num_tasks = 1
dataset.data.y = dataset.data.y.to(torch.float)
return (dataset[dataset.train_idx], dataset[dataset.val_idx],
dataset[dataset.test_idx]
), "fixed", MeanAbsoluteError(), "l1reg" #
elif name.startswith("count_cycle"):
sname = name[len("count_cycle"):]
yidx = int(sname)
trn_ds = dataset_random_graph("count_cycle", split="train", yidx=yidx)
val_ds = dataset_random_graph("count_cycle", split="val", yidx=yidx)
y_train_val = torch.cat([trn_ds.y, val_ds.y], dim=0)
mean = y_train_val.mean()
std = y_train_val.std()
trn_ds = dataset_random_graph("count_cycle",
split="train",
yidx=yidx,
ymean=mean,
ystd=std)
val_ds = dataset_random_graph("count_cycle",
split="val",
yidx=yidx,
ymean=mean,
ystd=std)
tst_ds = dataset_random_graph("count_cycle",
split="test",
yidx=yidx,
ymean=mean,
ystd=std)
trn_ds.num_tasks = 1
val_ds.num_tasks = 1
tst_ds.num_tasks = 1
# print(trn_ds.data.y.shape, trn_ds[0])
return (trn_ds, val_ds,
tst_ds), "fixed", MeanAbsoluteError(), "nodesmoothl1reg" #
elif name.startswith("count_graphlet"):
sname = name[len("count_graphlet"):]
yidx = int(sname)
trn_ds = dataset_random_graph("count_graphlet",
split="train",
yidx=yidx)
val_ds = dataset_random_graph("count_graphlet", split="val", yidx=yidx)
y_train_val = torch.cat([trn_ds.y, val_ds.y], dim=0)
mean = y_train_val.mean()
std = y_train_val.std()
trn_ds = dataset_random_graph("count_graphlet",
split="train",
yidx=yidx,
ymean=mean,
ystd=std)
val_ds = dataset_random_graph("count_graphlet",
split="val",
yidx=yidx,
ymean=mean,
ystd=std)
tst_ds = dataset_random_graph("count_graphlet",
split="test",
yidx=yidx,
ymean=mean,
ystd=std)
trn_ds.num_tasks = 1
val_ds.num_tasks = 1
tst_ds.num_tasks = 1
return (trn_ds, val_ds,
tst_ds), "fixed", MeanAbsoluteError(), "nodel1reg" #
elif name == "pascalvocsp":
xmean = torch.tensor([
4.2845e-01, 3.7611e-01, 1.4307e-01, 2.6746e-02, 3.0037e-02,
2.7267e-02, 5.0544e-01, 4.6751e-01, 2.3912e-01, 3.5321e-01,
2.8807e-01, 7.9090e-02, 1.9030e+02, 2.4771e+02
]) # mean of training data's x
xstd = torch.tensor([
2.5953e-01, 2.5717e-01, 2.7131e-01, 5.4823e-02, 5.4429e-02,
5.4475e-02, 2.6238e-01, 2.6601e-01, 2.7751e-01, 2.5197e-01,
2.4986e-01, 2.6070e-01, 1.1768e+02, 1.4007e+02
]) # std of training data's x
eamean = torch.tensor([0.0764,
33.7348]) # mean of training data's edge_attr
eastd = torch.tensor([0.0869,
20.9451]) # std of training data's edge_attr
def pascalvocsp_pre_transform(data):
data.x = (data.x - xmean) / xstd
data.edge_attr = (data.edge_attr - eamean) / eastd
return data
trn_ds = LRGBDataset("./dataset",
"PascalVOC-SP",
"train",
pre_transform=pascalvocsp_pre_transform)
val_ds = LRGBDataset("./dataset",
"PascalVOC-SP",
"val",
pre_transform=pascalvocsp_pre_transform)
tst_ds = LRGBDataset("./dataset",
"PascalVOC-SP",
"test",
pre_transform=pascalvocsp_pre_transform)
trn_ds.num_tasks = 21
val_ds.num_tasks = 21
tst_ds.num_tasks = 21
return (trn_ds, val_ds,
tst_ds), "fixed", F1Score("multiclass",
num_classes=21,
average="macro"), "nodecls" #
elif name == "pepfunc":
def pepfunc_pre_process(data):
data.x = data.x + 1
data.edge_attr = data.edge_attr + 1
return data
trn_ds = LRGBDataset("./dataset",
"Peptides-func",
"train",
pre_transform=pepfunc_pre_process).shuffle()
val_ds = LRGBDataset("./dataset",
"Peptides-func",
"val",
pre_transform=pepfunc_pre_process).shuffle()
tst_ds = LRGBDataset("./dataset",
"Peptides-func",
"test",
pre_transform=pepfunc_pre_process).shuffle()
trn_ds.num_tasks = 10
val_ds.num_tasks = 10
tst_ds.num_tasks = 10
return (trn_ds, val_ds,
tst_ds), "fixed", lambda x, y: MultilabelAveragePrecision(
10, average="macro")(x, (y > 0.5).to(torch.long)), "bincls"
elif name == "pepstruct":
def pepstruct_pre_process(data):
data.x = data.x + 1
data.edge_attr = data.edge_attr + 1
return data
trn_ds = LRGBDataset("./dataset",
"Peptides-struct",
"train",
pre_transform=pepstruct_pre_process)
val_ds = LRGBDataset("./dataset",
"Peptides-struct",
"val",
pre_transform=pepstruct_pre_process)
tst_ds = LRGBDataset("./dataset",
"Peptides-struct",
"test",
pre_transform=pepstruct_pre_process)
trn_ds.num_tasks = 11
val_ds.num_tasks = 11
tst_ds.num_tasks = 11
return (trn_ds, val_ds, tst_ds), "fixed", MeanAbsoluteError(), "l1reg"
elif name in ["MUTAG", "DD", "PROTEINS", "PTC_MR", "IMDB-BINARY"]:
def TUpretransform(data):
if data.x is None:
data.x = torch.ones((data.num_nodes, 1), dtype=torch.float)
else:
data.x = data.x + 1
if data.edge_attr is None:
data.edge_attr = torch.ones((data.edge_index.shape[1]), dtype=torch.float)
else:
data.edge_attr = data.edge_attr + 1
data.y = data.y.reshape(1, 1).to(torch.float)
return data
dataset = TUDataset("dataset", name=name, use_edge_attr=True, use_node_attr=True, pre_transform=TUpretransform, **kwargs)
dataset.num_tasks = 1
return (dataset, ), "fold-9-1-0", Accuracy("binary"), "bincls"
elif name == "zinc":
def zincpretransform(data):
data.x = data.x + 1
data.y = data.y.reshape(-1, 1)
data.edge_attr = (data.edge_attr + 1).to(torch.long).reshape(-1, 1)
return data
trn_d = ZINC("dataset/ZINC",
subset=True,
split="train",
pre_transform=zincpretransform,
**kwargs)
val_d = ZINC("dataset/ZINC",
subset=True,
split="val",
pre_transform=zincpretransform)
tst_d = ZINC("dataset/ZINC",
subset=True,
split="test",
pre_transform=zincpretransform)
trn_d.num_tasks = 1
val_d.num_tasks = 1
tst_d.num_tasks = 1
return (trn_d, val_d,
tst_d), "fixed", MeanAbsoluteError(), "smoothl1reg" #"reg"
elif name == "zinc-full":
def zincpretransform(data):
data.x = data.x + 1
data.y = data.y.reshape(-1, 1)
data.edge_attr = (data.edge_attr + 1).to(torch.long).reshape(-1, 1)
return data
trn_d = ZINC("dataset/ZINC",
subset=False,
split="train",
pre_transform=zincpretransform,
**kwargs)
val_d = ZINC("dataset/ZINC",
subset=False,
split="val",
pre_transform=zincpretransform)
tst_d = ZINC("dataset/ZINC",
subset=False,
split="test",
pre_transform=zincpretransform)
trn_d.num_tasks = 1
val_d.num_tasks = 1
tst_d.num_tasks = 1
return (trn_d, val_d,
tst_d), "fixed", MeanAbsoluteError(), "smoothl1reg" #"reg"
elif name.startswith("qm9"):
y_slice = int(name[3:])
def qm9pretrans(data):
data.x = (data.x + 1)# .to(torch.long)
data.edge_attr = (data.edge_attr + 1)# .to(torch.long)
return data
def qm9trans(data):
data.y = data.y[:, [y_slice]]
return data
dataset = QM9("dataset/qm9", transform=qm9trans, pre_transform=qm9pretrans, **kwargs)
tenpercent = int(len(dataset) * 0.1)
dataset.num_tasks = 1
dataset = dataset.shuffle()
test_dataset = dataset[:tenpercent]
val_dataset = dataset[tenpercent:2 * tenpercent]
train_dataset = dataset[2 * tenpercent:]
return (train_dataset, val_dataset,
test_dataset), "8-1-1", MeanAbsoluteError(), "l1reg"
elif name.startswith("ogbg"):
def pretransform(data):
data.edge_attr = data.edge_attr + 1
return data
dataset = PygGraphPropPredDataset(name=name,
pre_transform=pretransform)
split_idx = dataset.get_idx_split()
if "molhiv" in name:
task = "bincls"
elif "pcba" in name:
task = "bincls"
else:
raise NotImplementedError
dataset.y = dataset.y.to(torch.float)
dataset.x = dataset.x + 1
return (dataset[split_idx["train"]], dataset[split_idx["valid"]],
dataset[split_idx["test"]]), "fixed", myEvaluator(name), task
else:
raise NotImplementedError(name)
if __name__ == "__main__":
datalist = [
"IMDB-BINARY"#"PROTEINS", "DD","PROTEINS","IMDB-BINARY", "PTC_MR", "MUTAG"
] # "DD","PROTEINS","IMDB-BINARY", "PTC_MR", "MUTAG", "QM9", "IMDB-BINARY","sr", "EXP", "CSL", "subgcount0", "zinc", "ogbg-molhiv", "ogbg-molpcba"
for ds in datalist:
datasets = loaddataset(ds)[0]
dataset = datasets[0]
data = dataset[0]
print(ds, max([g.num_nodes for g in dataset]), len(dataset), data.x, data.edge_attr)#, data.x, data.y)
'''
PROTEINS 620 1113 Data(edge_index=[2, 162], x=[42, 4], y=[1, 1], edge_attr=[162])
DD 5748 1178 Data(edge_index=[2, 1798], x=[327, 89], y=[1, 1], edge_attr=[1798])
PROTEINS 620 1113 Data(edge_index=[2, 162], x=[42, 4], y=[1, 1], edge_attr=[162])
IMDB-BINARY 136 1000 Data(edge_index=[2, 146], y=[1, 1], x=[20, 1], edge_attr=[146], num_nodes=20)
PTC_MR 64 344 Data(edge_index=[2, 2], x=[2, 18], edge_attr=[2, 4], y=[1, 1])
MUTAG 28 188 Data(edge_index=[2, 38], x=[17, 7], edge_attr=[38, 4], y=[1, 1])
'''