-
Notifications
You must be signed in to change notification settings - Fork 1
/
datasets.py
432 lines (356 loc) · 14.1 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
import torch
import numpy as np
import networkx as nx
from torch_geometric.utils import from_networkx
from torch_geometric.utils import to_networkx
from scipy.spatial import Delaunay
class PrefixSumK():
# Creates a PrefixSum on a path - output must be sum mod 2
# input is 2 values [value, isRoot]
def __init__(self, k = 2, inp = 2):
super().__init__()
self.num_classes = k
self.num_features = 2
self.name = "PrefixSum mod K"
self.inp = inp
def gen_graph(self, s):
n = len(s)
rand_perm = np.arange(n)
G = nx.path_graph(rand_perm)
leafs = np.random.permutation([x for x in G.nodes() if G.degree(x)==1])
root = rand_perm[0]
labels = [[0.0, 0.0] for i in range(n)]
ylabels = [[0.0] for i in range(n)]
labels[root] = [0.0, 1.0]
counter = 0
for i,node in enumerate(rand_perm):
x = int(s[i])
labels[node][0] = s[i]
counter = (counter+ x)%self.num_classes
ylabels[node] = counter
dG = from_networkx(G)
dG.y = torch.tensor(ylabels)
dG.x = torch.tensor(labels)
return dG
def makedata(self, num_graphs = 200, num_nodes = 8, allow_sizes = False):
binary_strs = []
while len(binary_strs) < num_graphs:
graph_size = num_nodes
if allow_sizes:
graph_size = np.random.randint(2, graph_size+1)
ss = [np.random.randint(0,self.inp)*1.0 for _ in range(graph_size)]
if ss not in binary_strs:
binary_strs.append(ss)
return [self.gen_graph(s) for s in binary_strs]
class PrefixSum():
# Creates a PrefixSum on a path - output must be sum mod 2
# input is one-hot [value, isRoot]
def __init__(self):
super().__init__()
self.num_classes = 2
self.num_features = 4
self.name = "PrefixSum"
def gen_graph(self, s):
n = len(s)
rand_perm = np.arange(n)
G = nx.path_graph(rand_perm)
leafs = np.random.permutation([x for x in G.nodes() if G.degree(x)==1])
root = rand_perm[0]
labels = [[0.0, 0.0, 1.0, 0.0] for i in range(n)]
ylabels = [[0.0] for i in range(n)]
labels[root] = [0.0, 0.0, 0.0, 1.0]
counter = 0
for i,node in enumerate(rand_perm):
x = int(s[i])
labels[node][x] = 1.0
counter = (counter+ x)%2
ylabels[node] = counter
dG = from_networkx(G)
dG.y = torch.tensor(ylabels)
dG.x = torch.tensor(labels)
dG.edge_attr = torch.ones(G.number_of_edges()*2, 1)
return dG
def makedata(self, num_graphs = 200, num_nodes = 8, allow_sizes = False):
binary_strs = []
while len(binary_strs) < num_graphs:
graph_size = num_nodes
if allow_sizes:
graph_size = np.random.randint(2, graph_size+1)
ss = ''.join([str(np.random.randint(0,2)) for _ in range(graph_size)])
if ss not in binary_strs:
binary_strs.append(ss)
return [self.gen_graph(s) for s in binary_strs]
class Trees():
# Creates a Tree and marks the shortest path between two nodes
# input is one-hot [isEndpoint]
def __init__(self):
super().__init__()
self.num_classes = 2
self.num_features = 2
self.name = "ShortestPathTrees"
def gen_graph(self, num_nodes, num):
nx_graph = nx.random_tree(n=num_nodes, seed=num)
tree = from_networkx(nx_graph)
tree.x = torch.zeros(num_nodes, 2)
tree.y = torch.zeros(num_nodes)
tree.x[0][1] = 1
tree.x[1][1] = 1
shortest_path = nx.shortest_path(nx_graph, source=0, target=1)
for node in shortest_path:
tree.y[node] = 1
for node in range(num_nodes):
if tree.x[node][1] == 0:
tree.x[node][0] = 1
tree.edge_attr = torch.ones(nx_graph.number_of_edges()*2, 1)
return tree
def makedata(self, num_graphs = 200, num_nodes = 8, allow_sizes = False):
return [self.gen_graph(num_nodes, i) for i in range(num_graphs)]
class MidPoint():
def __init__(self):
super().__init__()
self.num_classes = 2
self.num_features = 2
self.name = "MidPoint"
def gen_graph(self, graph_size, posA, posB):
n = graph_size
rand_perm = np.arange(n)
G = nx.path_graph(rand_perm)
leafs = np.random.permutation([x for x in G.nodes() if G.degree(x)==1])
labels = [[1.0, 0.0] for i in range(n)]
ylabels = [0.0 for i in range(n)]
labels[posA] = [0.0, 1.0]
labels[posB] = [0.0, 1.0]
ylabels[int((posA+posB)/2)] = 1.0
dG = from_networkx(G)
dG.y = torch.tensor(ylabels)
dG.x = torch.tensor(labels)
return dG
def makedata(self, num_graphs = 200, num_nodes = 8, allow_sizes = False):
graph_str = []
graph_list = []
while len(graph_str) < num_graphs:
graph_size = num_nodes
if allow_sizes:
graph_size = np.random.randint(graph_size - 5, graph_size+1)
posA, posB = 0,0
for j in range(1000):
posA = np.random.randint(0, graph_size)
posB = np.random.randint(0, graph_size)
if 0 <= posA and posA < graph_size and posA < posB and posB < graph_size and (posB - posA)%2 == 0:
success = True
break
ss = f"{graph_size},{posA},{posB}"
if not success or ss in graph_str:
continue
graph_str.append(ss)
graph_list.append(self.gen_graph(graph_size, posA, posB))
return graph_list
class Cycles():
def __init__(self):
super().__init__()
self.num_classes = 2
self.num_features = 4
self.name = "Cycles"
def gen_graph(self, graph_size, posA, posB):
n = graph_size
rand_perm = np.arange(n)
G = nx.path_graph(rand_perm)
leafs = np.random.permutation([x for x in G.nodes() if G.degree(x)==1])
G.add_edge(leafs[0], leafs[1])
labels = [[1.0, 0.0] for i in range(n)]
ylabels = [0.0 for i in range(n)]
labels[posA] = [0.0, 1.0]
labels[posB] = [0.0, 1.0]
dist = posB - posA
odist = (graph_size - dist)//2
ylabels[(posA + posB)//2] = 1.0
ylabels[(posA-odist+graph_size)%graph_size] = 1.0
dG = from_networkx(G)
dG.y = torch.tensor(ylabels)
dG.x = torch.tensor(labels)
return dG
def makedata(self, num_graphs=10, num_nodes=8, allow_sizes=False):
graph_str = []
graph_list = []
while len(graph_str) < num_graphs:
graph_size = num_nodes
if allow_sizes:
graph_size = np.random.randint(4, graph_size+1)
posA, posB = 0,0
succ = False
for i in range(100):
posA = 0
posB = posA + np.random.randint(1, graph_size-posA)
if 0 <= posA and posA < graph_size and posA < posB and posB < graph_size:
if posB - posA <= graph_size//2 and (posB - posA)%2 == 0:
succ = True
break
ss = f"{graph_size},{posB-posA}"
if not succ or ss in graph_str:
continue
graph_str.append(ss)
graph_list.append(self.gen_graph(graph_size, posA, posB))
return graph_list
def randomgraph(n, **args):
g = nx.Graph()
g.add_nodes_from(range(n))
tree = set()
nodes = list(range(n))
current = np.random.choice(nodes)
tree.add(current)
while(len(tree) < n):
nxt = np.random.choice(nodes)
if not nxt in tree:
tree.add(nxt)
g.add_edge(current, nxt)
g.add_edge(nxt, current)
current = nxt
for _ in range(n//5):
i, j = np.random.permutation(n)[:2]
while g.has_edge(i,j):
i, j = np.random.permutation(n)[:2]
g.add_edge(i, j)
g.add_edge(j, i)
return g
def get_localized_distances(g, n):
seen = set()
distances = {}
queue = [(n, 0)]
while queue:
node, distance = queue.pop(0)
if node in distances and distances[node] < distance:
continue
distances[node] = distance
for nb in g.neighbors(node):
if nb not in seen:
seen.add(node)
queue.append((nb, distance + 1))
return [distances[i] for i in range(g.number_of_nodes())]
class Distance():
def __init__(self, num_graphs=200, num_nodes=12):
super().__init__()
self.num_features = 2
self.num_classes = 2
self.name = "Distance"
def gen_graph(self, num_nodes):
g = randomgraph(num_nodes)
origin = np.random.randint(0, num_nodes)
queue = [(origin, 0)]
seen = {origin}
even = set()
while queue:
node, distance = queue.pop(0)
if distance % 2 == 0:
even.add(node)
for nb in g.neighbors(node):
if nb not in seen:
seen.add(nb)
queue.append((nb, distance + 1))
data = from_networkx(g)
data.x = torch.tensor([[1.0,0.0] if x != origin else [0.0,1.0] for x in range(num_nodes)])
#data.x[origin:origin+1,:] = torch.ones(1, self.num_features).float()
#data.x[origin] = torch.ones([0.0,1.0])
distances = get_localized_distances(g, origin)
#data.diameter = max(distances)
#data.distances = torch.tensor(distances).unsqueeze(1)
data.edge_attr = torch.ones(g.number_of_edges()*2, 1)
data.y = torch.tensor([0.0 if n in even else 1.0 for n in range(num_nodes)])
return data
def makedata(self, num_graphs = 200, num_nodes = 8, allow_sizes = False):
return [self.gen_graph(num_nodes) for _ in range(num_graphs)]
class DistanceK():
def __init__(self, k = 2):
super().__init__()
self.num_features = 2
self.num_classes = k
self.name = "Distance"
def gen_graph(self, num_nodes):
g = randomgraph(num_nodes)
origin = np.random.randint(0, num_nodes)
queue = [(origin, 0)]
seen = {origin}
even = set()
while queue:
node, distance = queue.pop(0)
if distance % 2 == 0:
even.add(node)
for nb in g.neighbors(node):
if nb not in seen:
seen.add(nb)
queue.append((nb, distance + 1))
data = from_networkx(g)
data.x = torch.tensor([[1.0,0.0] if x != origin else [0.0,1.0] for x in range(num_nodes)])
#data.x[origin:origin+1,:] = torch.ones(1, self.num_features).float()
#data.x[origin] = torch.ones([0.0,1.0])
#distances = get_localized_distances(g, origin)
distances = nx.shortest_path_length(g, origin)
#data.diameter = max(distances)
#data.distances = torch.tensor(distances).unsqueeze(1)
data.edge_attr = torch.ones(g.number_of_edges()*2, 1)
data.y = torch.tensor([distances[n]%self.num_classes for n in range(num_nodes)])
#print(torch.tensor([0.0 if n in even else 1.0 for n in range(num_nodes)]))
return data
def makedata(self, num_graphs = 200, num_nodes = 8, allow_sizes = False):
return [self.gen_graph(num_nodes) for _ in range(num_graphs)]
class Distance_Delaunay():
def __init__(self, num_graphs=200, num_nodes=12):
super().__init__()
self.num_features = 2
self.num_classes = 2
self.name = "Distance_Delaunay"
def gen_graph(self, num_nodes):
points = np.random.rand(num_nodes, 2)
triangulation = Delaunay(points).vertex_neighbor_vertices
G = nx.Graph()
for i in range(num_nodes):
G.add_node(i)
for i in range(len(triangulation[0])-1):
for j in range(triangulation[0][i], triangulation[0][i+1]):
G.add_edge(i, triangulation[1][j])
origin = np.random.randint(0, num_nodes)
data = from_networkx(G)
data.x = torch.tensor([[1.0,0.0] if x != origin else [0.0,1.0] for x in range(num_nodes)])
distances = nx.single_source_shortest_path_length(G, origin)
data.y = torch.tensor([distances[n]%2 for n in range(num_nodes)])
data.edge_attr = torch.ones(G.number_of_edges()*2, 1)
return data
def makedata(self, num_graphs = 200, num_nodes = 8, allow_sizes = False):
return [self.gen_graph(num_nodes) for _ in range(num_graphs)]
def check_solvability(dataset):
seen_hashes_per_label = {}
solvable = True
for graph in dataset:
graph_nx = to_networkx(graph, to_undirected=True, node_attrs=['x', 'y'])
graph_hash = nx.weisfeiler_lehman_subgraph_hashes(graph_nx, node_attr='x', iterations=graph_nx.number_of_nodes())
for node in graph_nx.nodes:
hash = graph_hash[node][-1]
node_label = graph_nx.nodes[node]['y']
if node_label in seen_hashes_per_label:
seen_hashes_per_label[node_label].add(hash)
else:
seen_hashes_per_label[node_label] = set()
seen_hashes_per_label[node_label].add(hash)
seen_hashes = set()
unique_labels = seen_hashes_per_label.keys()
for label in unique_labels:
if not solvable:
break
for hash in seen_hashes_per_label[label]:
if hash in seen_hashes:
solvable = False
break
else:
seen_hashes.add(hash)
return solvable
if __name__ == '__main__':
print("Welcome to DATA.PY -- generate some graphs?")
datasets = [PrefixSumK(5), PrefixSum(), Trees(), Distance(), DistanceK(5), Distance_Delaunay()]
for dataset in datasets:
print(dataset.name)
data = dataset.makedata(num_graphs=100, num_nodes=10)
solvable = check_solvability(data)
if solvable:
print('All good')
else:
print('Dataset not solvable!!!')
exit()