-
Notifications
You must be signed in to change notification settings - Fork 5
/
main.py
323 lines (279 loc) · 11.8 KB
/
main.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
import logger as logger
from params import args
from logger import log
from handler import *
from model import *
from utils import *
import numpy as np
import torch as t
import pickle
import sys
import os
t.manual_seed(args.seed)
np.random.seed(args.seed)
class Coach:
def __init__(self, handler):
self.handler = handler
log(f"Users: {args.user}, Items(+1): {args.item}")
self.metrics = dict()
mets = ['loss', 'loss_main', 'hr@10', 'ndcg@10']
for met in mets:
self.metrics['Train' + met] = list()
self.metrics['Test' + met] = list()
def make_print(self, name, ep, reses, save):
ret = 'Epoch %d/%d, %s: ' % (ep, args.epoch, name)
for metric in reses:
val = reses[metric]
ret += '%s = %.4f, ' % (metric, val)
tem = name + metric
if save and tem in self.metrics:
self.metrics[tem].append(val)
ret = ret[:-2] + ' '
return ret
def run(self):
self.prepare_model()
log('Model Prepared')
if args.load_model != None:
self.load_model()
stloc = len(self.metrics['Trainloss']) * args.test_frequency - (args.test_frequency - 1)
else:
stloc = 0
log('Model Initialized')
bestRes = None
reses = self.test_epoch()
for ep in range(stloc, args.epoch):
tst_flag = (ep % args.test_frequency == 0)
reses = self.train_epoch()
log(self.make_print('Train', ep, reses, tst_flag))
sys.stdout.flush()
if tst_flag:
reses = self.test_epoch()
log(self.make_print('Test', ep, reses, tst_flag))
sys.stdout.flush()
if bestRes is None or reses['hr@10'] > bestRes['hr@10']:
bestRes = reses
log(self.make_print('Best Result', args.epoch, bestRes, True), bold=True)
self.save_history()
print()
reses = self.test_epoch()
log(self.make_print('Test', args.epoch, reses, True))
log(self.make_print('Best Result', args.epoch, bestRes, True), bold=True)
def prepare_model(self):
self.encoder = Encoder().cuda()
self.decoder = Decoder().cuda()
self.recommender = SASRec().cuda()
self.masker = RandomMaskSubgraphs()
self.sampler = LocalGraph()
self.opt = t.optim.Adam(
[{"params": self.encoder.parameters()},
{"params": self.decoder.parameters()},
{"params": self.recommender.parameters()}],
lr=args.lr, weight_decay=0
)
def sample_pos_edges(self, masked_edges):
return masked_edges[t.randperm(masked_edges.shape[0])[:args.con_batch]]
def sample_neg_edges(self, pos, dok):
neg = []
for u, v in pos:
cu_neg = []
num_samp = args.num_reco_neg // 2
for i in range(num_samp):
while True:
v_neg = np.random.randint(1, args.item)
if (u, v_neg) not in dok:
break
cu_neg.append([u, v_neg])
for i in range(num_samp):
while True:
u_neg = np.random.randint(1, args.item)
if (u_neg, v) not in dok:
break
cu_neg.append([u_neg, v])
neg.append(cu_neg)
return t.Tensor(neg).long()
def train_epoch(self):
self.encoder.train()
self.decoder.train()
self.recommender.train()
self.masker.train()
self.sampler.train()
loss_his = []
ep_loss, ep_loss_main, ep_loss_reco, ep_loss_mask = 0, 0, 0, 0
trn_loader = self.handler.trn_loader
steps = trn_loader.dataset.__len__() // args.trn_batch
for i, batch_data in enumerate(trn_loader):
if i % args.mask_steps == 0:
sample_scr, candidates = self.sampler(self.handler.ii_adj_all_one, self.encoder.get_ego_embeds())
masked_adj, masked_edg = self.masker(self.handler.ii_adj, candidates)
batch_data = [i.cuda() for i in batch_data]
seq, pos, neg = batch_data
item_emb, item_emb_his = self.encoder(masked_adj)
seq_emb = self.recommender(seq, item_emb)
tar_msk = pos > 0
loss_main = cross_entropy(seq_emb, item_emb[pos], item_emb[neg], tar_msk)
pos = self.sample_pos_edges(masked_edg)
neg = self.sample_neg_edges(pos, self.handler.ii_dok)
loss_reco = self.decoder(item_emb_his, pos, neg)
loss_regu = (calc_reg_loss(self.encoder) + calc_reg_loss(self.decoder) + calc_reg_loss(self.recommender)) * args.reg
loss = loss_main + loss_reco + loss_regu
loss_his.append(loss_main)
if i % args.mask_steps == 0:
reward = calc_reward(loss_his, args.eps)
loss_mask = -sample_scr.mean() * reward
ep_loss_mask += loss_mask
loss_his = loss_his[-1:]
loss += loss_mask
ep_loss += loss.item()
ep_loss_main += loss_main.item()
ep_loss_reco += loss_reco.item()
log('Step %d/%d: loss = %.3f, loss_main = %.3f loss_regu = %.3f, loss_reco = %.3f ' % (i, steps, loss, loss_main, loss_regu, loss_reco), save=False, oneline=True)
sys.stdout.flush()
self.opt.zero_grad()
loss.backward()
self.opt.step()
ret = dict()
ret['loss'] = ep_loss / steps
ret['loss_main'] = ep_loss_main / steps
ret['loss_reco'] = ep_loss_reco / steps
ret['loss_mask'] = ep_loss_mask / (steps // args.mask_steps)
return ret
def test_epoch(self):
self.encoder.eval()
self.decoder.eval()
self.recommender.eval()
self.masker.eval()
self.sampler.eval()
tst_loader = self.handler.tst_loader
ep_h5, ep_n5, ep_h10, ep_n10, ep_h20, ep_n20, ep_h50, ep_n50 = [0] * 8
group_h20 = [0] * 4
group_n20 = [0] * 4
group_num = [0] * 4
num = tst_loader.dataset.__len__()
steps = num // args.tst_batch
with t.no_grad():
for i, batch_data in enumerate(tst_loader):
batch_data = [i.cuda() for i in batch_data]
seq, pos, neg = batch_data
item_emb, item_emb_his = self.encoder(self.handler.ii_adj)
seq_emb = self.recommender(seq, item_emb)
seq_emb = seq_emb[:,-1,:] # (batch, 1, latdim)
all_ids = t.cat([pos, neg], -1) # (batch, 100)
all_emb = item_emb[all_ids] # (batch, 100, latdim)
all_scr = t.sum(t.unsqueeze(seq_emb, 1) * all_emb, -1) # (batch, 100)
seq_len = (seq > 0).cpu().numpy().sum(-1)
h5, n5, h10, n10, h20, n20, h50, n50, gp_h20, gp_n20, gp_num= \
self.calc_res(all_scr.cpu().numpy(), all_ids.cpu().numpy(), pos.cpu().numpy(), seq_len)
ep_h5 += h5
ep_n5 += n5
ep_h10 += h10
ep_n10 += n10
ep_h20 += h20
ep_n20 += n20
ep_h50 += h50
ep_n50 += n50
for j in range(4):
group_h20[j] += gp_h20[j]
group_n20[j] += gp_n20[j]
group_num[j] += gp_num[j]
log('Steps %d/%d: hr@10 = %.2f, ndcg@10 = %.2f ' % (i, steps, h10, n10), save=False, oneline=True)
sys.stdout.flush()
ep_h5 /= num
ep_n5 /= num
ep_h10 /= num
ep_n10 /= num
ep_h20 /= num
ep_n20 /= num
ep_h50 /= num
ep_n50 /= num
for i in range(4):
group_h20[i] /= group_num[i]
group_n20[i] /= group_num[i]
ret = dict()
ret['hr@10'] = ep_h10
ret['ndcg@10'] = ep_n10
print(f'Test result: h5={ep_h5:.4f} n5={ep_n5:.4f} h10={ep_h10:.4f} n10={ep_n10:.4f} h20={ep_h20:.4f} n20={ep_n20:.4f} h50={ep_h50:.4f} n50={ep_n50:.4f}')
return ret
def calc_res(self, scores, tst_ids, pos_ids, seq_len):
group_h20 = [0] * 4
group_n20 = [0] * 4
group_num = [0] * 4
h5, n5, h10, n10, h20, n20, h50, n50 = [0] * 8
for i in range(len(pos_ids)):
ids_with_scores = list(zip(tst_ids[i], scores[i]))
ids_with_scores = sorted(ids_with_scores, key=lambda x: x[1], reverse=True)
if seq_len[i] < 5:
group_num[0] += 1
elif seq_len[i] >= 5 and seq_len[i] < 10:
group_num[1] += 1
elif seq_len[i] >= 10 and seq_len[i] < 20:
group_num[2] += 1
else:
group_num[3] += 1
shoot = list(map(lambda x: x[0], ids_with_scores[:5]))
if pos_ids[i] in shoot:
h5 += 1
n5 += np.reciprocal(np.log2(shoot.index(pos_ids[i]) + 2))
shoot = list(map(lambda x: x[0], ids_with_scores[:10]))
if pos_ids[i] in shoot:
h10 += 1
n10 += np.reciprocal(np.log2(shoot.index(pos_ids[i]) + 2))
shoot = list(map(lambda x: x[0], ids_with_scores[:20]))
if pos_ids[i] in shoot:
if seq_len[i] < 5:
group_h20[0] += 1
group_n20[0] += np.reciprocal(np.log2(shoot.index(pos_ids[i]) + 2))
elif seq_len[i] >= 5 and seq_len[i] < 10:
group_h20[1] += 1
group_n20[1] += np.reciprocal(np.log2(shoot.index(pos_ids[i]) + 2))
elif seq_len[i] >= 10 and seq_len[i] < 20:
group_h20[2] += 1
group_n20[2] += np.reciprocal(np.log2(shoot.index(pos_ids[i]) + 2))
else:
group_h20[3] += 1
group_n20[3] += np.reciprocal(np.log2(shoot.index(pos_ids[i]) + 2))
h20 += 1
n20 += np.reciprocal(np.log2(shoot.index(pos_ids[i]) + 2))
shoot = list(map(lambda x: x[0], ids_with_scores[:50]))
if pos_ids[i] in shoot:
h50 += 1
n50 += np.reciprocal(np.log2(shoot.index(pos_ids[i]) + 2))
return h5, n5, h10, n10, h20, n20, h50, n50, group_h20, group_n20, group_num
def save_history(self):
if args.epoch == 0:
return
if not os.path.exists('./Models/'):
os.makedirs('./Models/')
if not os.path.exists('./History/'):
os.makedirs('./History/')
with open('./History/' + args.save_path + '.his', 'wb') as fs:
pickle.dump(self.metrics, fs)
content = {
'encoder': self.encoder,
'decoder': self.decoder,
'recommender': self.recommender,
}
t.save(content, './Models/' + args.save_path + '.mod')
log('Model Saved: %s' % args.save_path)
def load_model(self):
ckp = t.load('./Models/' + args.load_model + '.mod')
self.encoder = ckp['encoder']
self.decoder= ckp['decoder']
self.recommender = ckp['recommender']
self.opt = t.optim.Adam(
[{"params": self.encoder.parameters()},
{"params": self.decoder.parameters()},
{"params": self.recommender.parameters()}],
lr=args.lr, weight_decay=0
)
with open('./History/' + args.load_model + '.his', 'rb') as fs:
self.metrics = pickle.load(fs)
log('Model Loaded from ' + args.load_model)
if __name__ == '__main__':
logger.saveDefault = True
print(args)
log('Start')
handler = DataHandler()
handler.load_data()
log('Load Data')
coach = Coach(handler)
coach.run()