-
Notifications
You must be signed in to change notification settings - Fork 2
/
train.py
488 lines (390 loc) · 17.2 KB
/
train.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
import numpy as np
import scipy.sparse as sp
import os
import torch
import argparse
import torch.nn as nn
from torch.nn.modules.module import Module
from torch.nn.parameter import Parameter
from sklearn.metrics.cluster import normalized_mutual_info_score as nmi_score
from sklearn.metrics import adjusted_rand_score as ari_score
from sklearn.cluster import KMeans
from utils import sparse_mx_to_torch_sparse_tensor
from dataset import load
from torch.optim import Adam
import torch.nn.functional as F
import random
import collections
from sklearn.cluster import DBSCAN
def get_negative_mask(batch_size):
negative_mask = torch.ones((batch_size, batch_size),
dtype=bool) # torch.ones((batch_size, 2 * batch_size), dtype=bool)
for i in range(batch_size):
negative_mask[i, i] = 0 # negative_mask[i, i:] = 0
# negative_mask[i, i + batch_size] = 0
# negative_mask = torch.cat((negative_mask, negative_mask), 0)
return negative_mask
def cluster_acc(y_true, y_pred):
"""
Calculate clustering accuracy. Require scikit-learn installed
# Arguments
y: true labels, numpy.array with shape `(n_samples,)`
y_pred: predicted labels, numpy.array with shape `(n_samples,)`
# Return
accuracy, in [0,1]
"""
y_true = y_true.astype(np.int64)
assert y_pred.size == y_true.size
D = max(y_pred.max(), y_true.max()) + 1
w = np.zeros((D, D), dtype=np.int64)
for i in range(y_pred.size):
w[y_pred[i], y_true[i]] += 1
from sklearn.utils.linear_assignment_ import linear_assignment
ind = linear_assignment(w.max() - w)
return sum([w[i, j] for i, j in ind]) * 1.0 / y_pred.size
def target_distribution(q):
weight = q ** 2 / q.sum(0)
return (weight.t() / weight.sum(1)).t()
# Borrowed from https://github.com/PetarV-/DGI
class GCN(nn.Module):
def __init__(self, in_ft, out_ft, bias=True):
super(GCN, self).__init__()
self.fc = nn.Linear(in_ft, out_ft, bias=False)
self.act = nn.PReLU()
if bias:
self.bias = nn.Parameter(torch.FloatTensor(out_ft))
self.bias.data.fill_(0.0)
else:
self.register_parameter('bias', None)
for m in self.modules():
self.weights_init(m)
def weights_init(self, m):
if isinstance(m, nn.Linear):
torch.nn.init.xavier_uniform_(m.weight.data)
if m.bias is not None:
m.bias.data.fill_(0.0)
# Shape of seq: (batch, nodes, features)
def forward(self, seq, adj, sparse=False):
seq_fts = self.fc(seq)
if sparse:
out = torch.unsqueeze(torch.spmm(adj, torch.squeeze(seq_fts, 0)), 0)
else:
out = torch.bmm(adj, seq_fts)
if self.bias is not None:
out += self.bias
return self.act(out)
class GraphConvolution(Module):
"""
Simple GCN layer, similar to https://arxiv.org/abs/1609.02907
"""
def __init__(self, in_features, out_features, dropout=0., act=F.relu):
super(GraphConvolution, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.dropout = dropout
self.act = act
self.weight = Parameter(torch.FloatTensor(in_features, out_features))
self.reset_parameters()
def reset_parameters(self):
torch.nn.init.xavier_uniform_(self.weight)
def forward(self, input, adj):
input = F.dropout(input, self.dropout, self.training)
support = torch.mm(input, self.weight)
output = torch.spmm(adj, support)
output = self.act(output)
return output
def __repr__(self):
return self.__class__.__name__ + ' (' \
+ str(self.in_features) + ' -> ' \
+ str(self.out_features) + ')'
class InnerProductDecoder(nn.Module):
"""Decoder for using inner product for prediction."""
def __init__(self, dropout, act=torch.sigmoid):
super(InnerProductDecoder, self).__init__()
self.dropout = dropout
self.act = act
def forward(self, z):
z = F.dropout(z, self.dropout, training=self.training)
adj = self.act(torch.mm(z, z.t()))
return adj
# Borrowed from https://github.com/PetarV-/DGI
class Readout(nn.Module):
def __init__(self):
super(Readout, self).__init__()
def forward(self, seq, msk):
if msk is None:
return torch.mean(seq, 1)
else:
msk = torch.unsqueeze(msk, -1)
return torch.mean(seq * msk, 1) / torch.sum(msk)
# Borrowed from https://github.com/PetarV-/DGI
class Discriminator(nn.Module):
def __init__(self, n_h):
super(Discriminator, self).__init__()
self.f_k = nn.Bilinear(n_h, n_h, 1)
for m in self.modules():
self.weights_init(m)
def weights_init(self, m):
if isinstance(m, nn.Bilinear):
torch.nn.init.xavier_uniform_(m.weight.data)
if m.bias is not None:
m.bias.data.fill_(0.0)
def forward(self, c1, c2, h1, h2, h3, h4, s_bias1=None, s_bias2=None): # ???
c_x1 = torch.unsqueeze(c1, 1)
c_x1 = c_x1.expand_as(h1).contiguous()
c_x2 = torch.unsqueeze(c2, 1)
c_x2 = c_x2.expand_as(h2).contiguous()
# positive
sc_1 = torch.squeeze(self.f_k(h2, c_x1), 2)
sc_2 = torch.squeeze(self.f_k(h1, c_x2), 2)
# negetive
sc_3 = torch.squeeze(self.f_k(h4, c_x1), 2)
sc_4 = torch.squeeze(self.f_k(h3, c_x2), 2)
logits = torch.cat((sc_1, sc_2, sc_3, sc_4), 1)
return logits
class Model(nn.Module):
def __init__(self, n_in, n_h, nb_classes):
super(Model, self).__init__()
self.gcn1 = GCN(n_in, n_h)
self.gcn2 = GCN(n_in, n_h)
self.read = Readout()
self.sigm = nn.Sigmoid()
self.disc = Discriminator(n_h)
self.fc = nn.Linear(n_h, nb_classes)
self.sigm_fc = nn.Sigmoid()
def forward(self, seq1, seq2, diff, sparse):
# contra
h_mask = self.gcn2(seq2, diff, sparse)
h_2 = self.gcn2(seq1, diff, sparse)
return h_mask[0].unsqueeze(0), h_2[0].unsqueeze(0)
def embed(self, seq1, diff, sparse):
h_2 = self.gcn2(seq1, diff, sparse)
return h_2[0].unsqueeze(0).detach()
class UCGL(nn.Module):
def __init__(self, n_in, n_h, nb_classes, n_clusters, pretrain_path=''): # pretrain_path should be added
super(UCGL, self).__init__()
self.alpha = 0.0001 # 0.000001#1.0
self.pretrain_path = pretrain_path
self.mvg = Model(n_in, n_h, nb_classes)
# cluster layer
self.cluster_layer = Parameter(torch.Tensor(n_clusters, n_h)) # n_clusters?
torch.nn.init.xavier_normal_(self.cluster_layer.data)
def pretrain(self, adj, diff, features, labels, idx_train, idx_val, idx_test):
self.mvg.load_state_dict(torch.load(self.pretrain_path))
print('load pretrained MVGRL from', self.pretrain_path)
def embed(self, seq, adj, diff, sparse):
h_1 = self.mvg.gcn1(seq, adj, sparse)
h = self.mvg.gcn2(seq, diff, sparse)
return ((h + h_1)).detach()
def forward(self, bf, mask_fts, bd, sparse):
h_mask, h = self.mvg(bf, mask_fts, bd, sparse)
# cluster
q = 1.0 / (1.0 + torch.sum(
torch.pow(h.reshape(-1, h.shape[2]).unsqueeze(1) - self.cluster_layer, 2),
2) / self.alpha) # h.reshape(-1,h.shape[2]).unsqueeze(1)-self.cluster_layer
q = q.pow((self.alpha + 1.0) / 2.0)
q = (q.t() / torch.sum(q, 1)).t()
return h_mask, h, q
class fc_layer(nn.Module):
def __init__(self, ft_in, nb_classes):
super(fc_layer, self).__init__()
self.fc = nn.Linear(ft_in, nb_classes)
self.sigm = nn.Sigmoid()
for m in self.modules():
self.weights_init(m)
def weights_init(self, m):
if isinstance(m, nn.Linear):
torch.nn.init.xavier_uniform_(m.weight.data)
if m.bias is not None:
m.bias.data.fill_(0.0)
def forward(self, seq):
ret = torch.log_softmax(self.fc(seq), dim=-1)
return ret
def train_ucgl(dataset):
nb_epochs = 1500
lr = 0.00005
sparse = False
adj, diff, features, labels, idx_train, idx_val, idx_test = load(dataset)
ft_size = features.shape[1]
nb_classes = np.unique(labels).shape[0]
sample_size = features.shape[0]
batch_size = 4
labels = torch.LongTensor(labels)
model_UCGL = UCGL(ft_size, args.hid_units, nb_classes, n_clusters=args.n_clusters, pretrain_path=args.pretrain_path)
optimizer = torch.optim.Adam(model_UCGL.parameters(), lr=lr, weight_decay=0.0)
if torch.cuda.is_available():
model_UCGL = model_UCGL.cuda()
labels = labels.cuda()
model_UCGL.pretrain(adj, diff, features, labels, idx_train, idx_val, idx_test)
if sparse:
adj = sparse_mx_to_torch_sparse_tensor(sp.coo_matrix(adj))
diff = sparse_mx_to_torch_sparse_tensor(sp.coo_matrix(diff))
features_array = features
diff_array = diff
features = torch.FloatTensor(features[np.newaxis])
adj = torch.FloatTensor(adj[np.newaxis])
diff = torch.FloatTensor(diff[np.newaxis])
features = features.cuda()
adj = adj.cuda()
diff = diff.cuda()
# obtain features of positive samples
features_mask = features
for i in range(features_mask.shape[1]):
idx = random.sample(range(1, features_mask.shape[2]), args.mask_num)
features_mask[0][i][idx] = 0
features_mask_array = np.array(features_mask.squeeze(0).cpu())
# cluster parameter initiate
h2 = model_UCGL.mvg.embed(features, diff, sparse)
kmeans = KMeans(n_clusters=args.n_clusters)
y_pred = kmeans.fit_predict(h2.data.squeeze().cpu().numpy())
model_UCGL.cluster_layer.data = torch.tensor(kmeans.cluster_centers_).to(device)
model_UCGL.train()
acc_clu = 0
kl_loss = 0
loss = 0
for epoch in range(nb_epochs):
idx = np.random.randint(0, adj.shape[-1] - sample_size + 1, batch_size)
ba, bd, bf, bf_mask = [], [], [], []
for i in idx:
bd.append(diff_array[i: i + sample_size, i: i + sample_size])
bf.append(features_array[i: i + sample_size])
bf_mask.append(features_mask_array[i: i + sample_size])
bd = np.array(bd).reshape(batch_size, sample_size, sample_size)
bf = np.array(bf).reshape(batch_size, sample_size, ft_size)
bf_mask = np.array(bf_mask).reshape(batch_size, sample_size, ft_size)
if sparse:
bd = sparse_mx_to_torch_sparse_tensor(sp.coo_matrix(bd))
else:
bd = torch.FloatTensor(bd)
bf = torch.FloatTensor(bf)
bf_mask = torch.FloatTensor(bf_mask)
if torch.cuda.is_available():
bf = bf.cuda()
bd = bd.cuda()
bf_mask = bf_mask.cuda()
if epoch % args.update_interval == 0:
_, _, tmp_q = model_UCGL(bf, bf_mask, bd, sparse)
# update target distribution p
tmp_q = tmp_q.data
p = target_distribution(tmp_q)
# evaluate clustering performance
y_pred = tmp_q.cpu().numpy().argmax(1)
acc = cluster_acc(np.array(labels.cpu()), y_pred)
nmi = nmi_score(np.array(labels.cpu()), y_pred)
ari = ari_score(np.array(labels.cpu()), y_pred)
if acc > acc_clu:
acc_clu = acc
nmi_clu = nmi
ari_clu = ari
torch.save(model_UCGL.state_dict(), args.model_path)
h_mask, h_2_sour, q = model_UCGL(bf, bf_mask, bd, sparse)
kl_loss = F.kl_div(q.log(), p)
temperature = 0.5
y_sam = torch.LongTensor(y_pred)
neg_size = 1000
class_sam = []
for m in range(np.max(y_pred) + 1):
class_del = torch.ones(int(sample_size), dtype=bool)
class_del[np.where(y_sam.cpu() == m)] = 0
class_neg = torch.arange(sample_size).masked_select(class_del)
neg_sam_id = random.sample(range(0, class_neg.shape[0]), int(neg_size))
class_sam.append(class_neg[neg_sam_id])
out = (h_2_sour).squeeze()
neg = torch.exp(torch.mm(out, out.t().contiguous()) / temperature)
neg_samp = torch.zeros(neg.shape[0], int(neg_size))
for n in range(np.max(y_pred) + 1):
neg_samp[np.where(y_sam.cpu() == n)] = neg.cpu().index_select(1, class_sam[n])[np.where(y_sam.cpu() == n)]
neg_samp = neg_samp.cuda()
Ng = neg_samp.sum(dim=-1)
pos_size = 10
class_sam_pos = []
for m in range(np.max(y_pred) + 1):
class_del = torch.ones(int(sample_size), dtype=bool)
class_del[np.where(y_sam.cpu() != m)] = 0
class_pos = torch.arange(sample_size).masked_select(class_del)
pos_sam_id = random.sample(range(0, class_pos.shape[0]), int(pos_size))
class_sam_pos.append(class_neg[pos_sam_id])
out = h_2_sour.squeeze()
pos = torch.exp(torch.mm(out, out.t().contiguous()))
pos_samp = torch.zeros(pos.shape[0], int(pos_size))
for n in range(np.max(y_pred) + 1):
pos_samp[np.where(y_sam.cpu() == n)] = pos.cpu().index_select(1, class_sam_pos[n])[
np.where(y_sam.cpu() == n)]
pos_samp = pos_samp.cuda()
pos = pos_samp.sum(dim=-1) + torch.diag(torch.exp(torch.mm(out, (h_mask.squeeze()).t().contiguous())))
node_contra_loss_2 = (- torch.log(pos / (pos + Ng))).mean()
loss = node_contra_loss_2 + args.beta * kl_loss
optimizer.zero_grad()
loss.backward()
optimizer.step()
if acc < 0.688:
break
print('CLustering Acc: {:.4f}'.format(acc_clu), ', nmi {:.4f}'.format(nmi_clu), ', ari {:.4f}'.format(ari_clu))
if args.classfication == 'True':
class_loss = nn.CrossEntropyLoss()
model_UCGL.load_state_dict(torch.load(args.model_path))
embeds = model_UCGL.embed(features, adj, diff, sparse)
train_embs = embeds[0, idx_train]
test_embs = embeds[0, idx_test]
train_lbls = labels[idx_train]
test_lbls = labels[idx_test]
accs = []
wd = 0.01 if dataset == 'citeseer' else 0.0
for _ in range(50):
acc_max = 0
classificate = fc_layer(args.hid_units, nb_classes)
classificate = classificate.cuda()
opt = torch.optim.Adam(classificate.parameters(), lr=5e-3, weight_decay=wd)
for _ in range(700):
classificate.train()
opt.zero_grad()
logits = classificate(train_embs)
loss = class_loss(logits, train_lbls)
loss.backward()
opt.step()
preds = torch.argmax(classificate(test_embs), dim=1)
acc = torch.sum(preds == test_lbls).float() / test_lbls.shape[0]
if acc > acc_max:
acc_max = acc
print('class acc:{}'.format(acc_max))
accs.append(acc_max * 100)
accs = torch.stack(accs)
print(accs.mean().item(), accs.std().item())
if __name__ == '__main__':
import warnings
warnings.filterwarnings("ignore")
torch.cuda.set_device(0)
parser = argparse.ArgumentParser(
description='train_ucgl',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--lr', type=float, default=0.001)
parser.add_argument('--n_clusters', default=6, type=int)
parser.add_argument('--batch_size', default=256, type=int)
parser.add_argument('--hid_units', default=220, type=int)
parser.add_argument('--hidden1', type=int, default=16, help='Number of units in hidden layer 1.')
parser.add_argument('--hidden2', type=int, default=16, help='Number of units in hidden layer 2.')
parser.add_argument('--dataset', type=str, default='pubmed')
parser.add_argument('--dropout', type=float, default=0., help='Dropout rate (1 - keep probability).')
parser.add_argument('--update_interval', type=int)
parser.add_argument('--tol', default=0.000, type=float)
parser.add_argument('--pretrain_path', default='pretrained_model_pubmed.pkl', type=str,
help='the pretrained MVGRL path') # pretrained on MVGRL(https://github.com/kavehhassani/mvgrl)
parser.add_argument('--beta', default=10e-4, type=float, help='coefficient of kl loss')
parser.add_argument('--model_path', type=str,
default='', help='the optimized model after training under debiased contrastive loss')
parser.add_argument('--classfication', type=str, default='True')
parser.add_argument('--mask_num', type=int)
args = parser.parse_args()
args.cuda = torch.cuda.is_available()
device = torch.device("cuda:0")
args.dataset = 'pubmed'
if args.dataset == 'cora':
args.n_clusters = 7
elif args.dataset == 'pubmed':
args.n_clusters = 3
args.model_path = os.path.join('/youpath/UCGL_' + args.dataset + '_' + str(args.hid_units) + '.pkl')
#mask_num 200
#[pos_size, neg_size,lr,update_interval]
#Cora [10, 1000, 5*10e-5, 2], Citesser [70, 1000, 7*10e-5, 1], pubmed [450, 7000, 5*10e-5, 2]
for __ in range(50): # For reproducing through multiple running times due to the randomness of contrastive sample selection in each iteration
train_ucgl(args.dataset)