-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
449 lines (391 loc) · 13 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
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
import numpy as np
import torch
import torch.nn.functional as F
from torch_geometric.datasets import Planetoid
from torch_geometric.data import ClusterData, Data
from numpy.random import randint, random
import pickle
import logging
logging.basicConfig(level=logging.INFO,
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)
OBJ_END = 'eofeofeof'
IP_PORT_BASE = 7000
ADDR = "addr"
# ip port
def reset_ipport():
for label in [''] + list(range(100)):
path = '{}/ip_oprt{}.pkl'.format(ADDR, label)
with open(path, 'wb') as f:
pickle.dump(('localhost', IP_PORT_BASE + 100), f)
return
def read_ipport():
for label in [''] + list(range(100)):
path = '{}/ip_oprt{}.pkl'.format(ADDR, label)
with open(path, 'rb') as f:
print(pickle.load(f))
return
def get_ip_port(id=''):
path = '{}/ip_oprt{}.pkl'.format(ADDR, id)
with open(path, 'rb') as f:
ans = pickle.load(f)
if ans[1] >= IP_PORT_BASE + 1000:
new_ip_port = (ans[0], IP_PORT_BASE)
else:
new_ip_port = (ans[0], ans[1] + 100)
with open(path, 'wb') as f:
pickle.dump(new_ip_port, f)
return ans
# net contact
def socket_recv_with_response(socket):
'''
recv message with specified socket and response okk
:param socket:
:return: decoded data
'''
data = b''
while True:
packet = socket.recv(1024*4)
data += packet
if data[-9:].__contains__(OBJ_END.encode()):
data = data[:-9]
break
socket.send('okk'.encode())
if data.__sizeof__() >= 10*1024*1024:
print('recv big obj:{:.2f}M'.format(data.__sizeof__()/(1024*1024)))
return pickle.loads(data)
def socket_send_with_waiting_response(socket, mess):
'''
send out message and waiting for okk as response
:param socket:
:param mess:
:return:
'''
mess = pickle.dumps(mess)
socket.send(mess+OBJ_END.encode())
socket.recv(1024)
return
def socket_recv(socket):
'''
only recv message
not send okk
:param socket:
:return:
'''
data = b''
while True:
packet = socket.recv(1024*4)
data += packet
if data[-9:].__contains__(OBJ_END.encode()):
data = data[:-9]
break
if data.__sizeof__() >= 10*1024*1024:
print('recv big obj:{:.2f}M'.format(data.__sizeof__()/(1024*1024)))
return pickle.loads(data)
def socket_send(socket, mess):
'''
only send message
not waiting for response
:param socket:
:param mess:
:return:
'''
mess = pickle.dumps(mess)
socket.send(mess+OBJ_END.encode())
return
class Contacter():
'''
use this class as a interface to process data recv and send
'''
def __init__(self):
self.socket = None
return
def recv_with_res(self):
data = socket_recv_with_response(self.socket)
return data
def recv(self):
data = socket_recv(self.socket)
return data
def send_with_waiting_res(self, mess):
socket_send_with_waiting_response(self.socket, mess)
return
def send(self, mess):
socket_send(self.socket, mess)
return
# serealize data
def setalize(supermasks):
'''
remove the same supermask, ank keep only one copy for each supermask
:param supermasks:
:return:
'''
sm_set = set()
for tmp in supermasks:
sm_set.add(str(tmp))
setlized_sms = []
for tmp in sm_set:
setlized_sms.append(eval(tmp))
return setlized_sms
def serialize_model(model):
'''
encode all the parameters in a nn model
:param model: nn.Module
:return: dic
'''
param_dict = {}
for name, param in model.named_parameters():
param_dict[name] = param.to('cpu')
# print(param_dict)
return param_dict
# partition datasets
class PartitionTool():
def __init__(self):
return
def partition_subgraph(self, data, k, copy_node=False):
'''
partition a data into k parts
:param data: torch_geometric.data.Data
:param k:
:param copy_node:
:return:
'''
cluster = ClusterData(data, k)
print(len(cluster.perm))
print(cluster.partptr)
for i in range(k):
exec('self.idx_{} = cluster.perm[cluster.partptr[{}]:cluster.partptr[{}]]'.format(
i, i, i + 1))
print('cluster over!')
# expand
if copy_node == True:
def expand(edge_index, idx):
new_edge_index_u = []
new_edge_index_v = []
extra_node = set()
for tmp in idx:
extra_node.add(int(tmp))
for v in v2e[int(tmp)]:
extra_node.add(v)
new_edge_index_u += [int(tmp)
for i in range(len(v2e[int(tmp)]))]
new_edge_index_v += v2e[int(tmp)]
# print(len(id_set))
print('over for loop')
return torch.LongTensor(list(extra_node)), torch.LongTensor([new_edge_index_u, new_edge_index_v])
else:
def expand(edge_index, idx):
new_edge_index_u = []
new_edge_index_v = []
id_set = set()
for tmp in idx:
id_set.add(int(tmp))
for tmp in idx:
for v in v2e[int(tmp)]:
if id_set.__contains__(v):
new_edge_index_u.append(int(tmp))
new_edge_index_v.append(v)
return torch.LongTensor(list(idx)), torch.LongTensor([new_edge_index_u, new_edge_index_v])
edge_index = np.array(data.edge_index)
v2e = {}
# for i in range(len(edge_index[0])):
# u, v = edge_index[0][i], edge_index[1][i]
# v2e[u] = []
# v2e[v] = []
for u in range(len(data.x)):
v2e[u] = []
for i in range(len(edge_index[0])):
u, v = edge_index[0][i], edge_index[1][i]
v2e[u].append(v)
for i in range(k):
print(i)
exec('self.expanded_idx_{}, self.edge_index_{} = expand(edge_index, self.idx_{})'.format(
i, i, i))
print('subgraph expands over')
def hashing(idx, edge_index):
idx = np.array(idx)
edge_index = np.array(edge_index)
idx = np.sort(idx)
idx_map = np.zeros(int(np.max(idx)) + 1)
for i in range(len(idx)):
idx_map[int(idx[i])] = i
# print(idx_map)
for i in range(len(edge_index[0])):
edge_index[0][i], edge_index[1][i] = idx_map[int(
edge_index[0][i])], idx_map[int(edge_index[1][i])]
return torch.LongTensor(edge_index)
def idx_map(idx):
idx = np.array(idx)
idx = np.sort(idx)
ret_idx_map = np.zeros(int(np.max(idx)) + 1, dtype=int)
for i in range(len(idx)):
ret_idx_map[int(idx[i])] = i
return ret_idx_map
for i in range(k):
print(i)
exec(
'self.edge_index_{} = hashing(self.expanded_idx_{},self.edge_index_{})'.format(i, i, i))
print('hash over')
x = data.x
y = data.y
train_mask = data.train_mask
val_mask = data.val_mask
test_mask = data.test_mask
datas = []
train = 0
val = 0
test = 0
for i in range(k):
print(i)
exec('idx_{}_bool_tensor = torch.zeros(len(x)).bool()'.format(i))
exec(
'for idx in self.expanded_idx_{}:idx_{}_bool_tensor[idx] = True'.format(i, i))
exec('x_{} = x[idx_{}_bool_tensor]'.format(i, i))
exec('y_{} = y[idx_{}_bool_tensor]'.format(i, i))
exec(
'train_mask_{} = torch.zeros(len(self.expanded_idx_{})).bool()'.format(i, i))
exec('val_mask_{} = torch.zeros(len(self.expanded_idx_{})).bool()'.format(i, i))
exec(
'test_mask_{} = torch.zeros(len(self.expanded_idx_{})).bool()'.format(i, i))
ret_idx_map = idx_map(eval('self.expanded_idx_{}'.format(i)))
for idx in eval('self.idx_{}'.format(i)):
exec(
'train_mask_{}[ret_idx_map[idx]] = train_mask[idx]'.format(i))
exec('val_mask_{}[ret_idx_map[idx]] = val_mask[idx]'.format(i))
exec(
'test_mask_{}[ret_idx_map[idx]] = test_mask[idx]'.format(i))
train += eval('train_mask_{}'.format(i)).sum()
val += eval('val_mask_{}'.format(i)).sum()
test += eval('test_mask_{}'.format(i)).sum()
exec('data_{} = Data(x=x_{},y=y_{},edge_index=self.edge_index_{},train_mask=train_mask_{},val_mask=val_mask_{},test_mask=test_mask_{})'.format(
i, i, i, i, i, i, i))
datas.append(eval('data_{}'.format(i)))
print('sub graph pack over', train, val, test)
return datas
# cal edge_feat
def cal_edge_attr_for_gmmconv(edge_index):
'''
contruct edge_attr on specified dataset for gmmconv
:param edge_index:
:return:
'''
embedding_matrix = torch.bincount(
edge_index[0], minlength=edge_index.max()+1).unsqueeze(1)+1
edge_attr = F.embedding(
edge_index, embedding_matrix).pow(-0.5).T.squeeze(0)
return edge_attr.to(edge_index.device)
# evo related
MASKRANGE = [[1, 1], [1, 12], [13, 24], [
1, 36], [1, 48], [1, 60], [1, 72], [1, 5]]
def rand_element(idx):
zero_p = [0, 0, 0, 0.2, 0.2, 0.2, 0.2, 0]
rand_seed = zero_p[idx]
if random() < rand_seed:
return 0
return randint(MASKRANGE[idx][0], MASKRANGE[idx][1] + 1)
def random_supermask():
'''
generate random supermask
:return:
'''
mask = []
idx = 0
for r in MASKRANGE:
mask.append(rand_element(idx))
idx += 1
return mask
def cross_over(original, target):
son = []
for i in range(len(MASKRANGE)):
if np.random.random() > 0.5:
son.append(target[i])
else:
son.append(original[i])
return son
def mutate(original):
c = 0.125
son = []
for i, range in enumerate(MASKRANGE):
if np.random.random() < c:
son.append(rand_element(i))
else:
son.append(original[i])
return son
# metrics
def accuracy(output, labels):
'''
get accuracy
:param output:
:param labels:
:return:
'''
preds = output.max(1)[1].type_as(labels)
correct = preds.eq(labels).double()
correct = correct.sum().item()
return correct / len(labels)
def num_correct(output, labels):
'''
get the number of correct nodes that has been predicted correctly
:param output:
:param labels:
:return:
'''
preds = output.max(1)[1].type_as(labels)
correct = preds.eq(labels).double()
correct = correct.sum().item()
return int(correct)
def num_params(model):
params = list(model.named_parameters())
k = 0
for _, i in params:
l = 1
for j in i.size():
l *= j
k = k + l
return k
def parse_path(supermask):
def get_leaf_idx(supermask):
local_mask = torch.ones(7, dtype=int)
for i in range(1, 7):
local_mask[int((supermask[i]-1)/12)] = 0
if supermask[i] == 0:
local_mask[i] = 0
return local_mask.nonzero()
leaf_idx = get_leaf_idx(supermask)
def get_supermask_from_leaf(leaf_idx, supermask, ret=torch.zeros(len(supermask), dtype=int)):
ret[0], ret[-1] = supermask[0], supermask[-1]
idx = leaf_idx
code = supermask[idx]
while code and idx:
ret[idx] = code
idx = int((code-1)/12)
code = supermask[idx]
return ret
leaf_idx = sorted(leaf_idx, key=lambda x: get_supermask_from_leaf(
x, supermask, torch.zeros(len(supermask), dtype=int)).tolist().__str__(), reverse=True)
ans = []
for idx in leaf_idx:
if len(ans) < 0:
ans.append(get_supermask_from_leaf(
idx, supermask, ans[-1].clone()))
else:
ans.append(get_supermask_from_leaf(idx, supermask,
torch.zeros(len(supermask), dtype=int)))
return ans
if __name__ == "__main__":
# # partitioner = PartitionTool()
# # partitioner.partition_subgraph()
# for i in range(3):
# with open('data/citeseer/{}_{}copynode.pkl'.format(i, ''), 'rb') as f:
# data = pickle.load(f)
# print(data.edge_index.max(), data, data.train_mask.sum(),
# data.val_mask.sum(), data.test_mask.sum())
# # print(read_ipport())
# # x = torch.rand(5, 1)
# # y = torch.ones(5, 1)
# # print(x, x.where(x > 0.5, y))
import os
for id in range(2200, 2300):
os.system('kill -9 {}'.format(id))
# print(parse_path([1, 6, 21, 24, 43, 21, 49, 5]))
# print(parse_path([4, 12, 13, 6, 0, 42, 0, 2]))
pass