-
Notifications
You must be signed in to change notification settings - Fork 0
/
retraining-kdeep.py
270 lines (230 loc) · 11.1 KB
/
retraining-kdeep.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
import pickle
import argparse
import pandas as pd
import numpy as np
from tqdm.auto import tqdm
import torch
import torch.nn as nn
#import torchtext
import torchtext.legacy as torchtext
from torch.utils.data import DataLoader
import layers
import sampler as sampler_module
import evaluation
from ranger import Ranger
from scipy import spatial
from sklearn.neighbors import NearestNeighbors
class PinSAGEModel(nn.Module):
def __init__(self, full_graph, ntype, textsets, hidden_dims, n_layers):
super().__init__()
self.proj = layers.LinearProjector(full_graph, ntype, textsets, hidden_dims)
self.sage = layers.SAGENet(hidden_dims, n_layers)
self.scorer = layers.ItemToItemScorer(full_graph, ntype)
def forward(self, pos_graph, neg_graph, blocks):
h_item = self.get_repr(blocks)
pos_score = self.scorer(pos_graph, h_item)
neg_score = self.scorer(neg_graph, h_item)
return (neg_score - pos_score + 1).clamp(min=0)
def get_repr(self, blocks):
h_item = self.proj(blocks[0].srcdata)
h_item_dst = self.proj(blocks[-1].dstdata)
return h_item_dst + self.sage(blocks, h_item)
def prepare_dataset(data_dict, args):
g = data_dict['graph']
item_texts = data_dict['item_texts']
user_ntype = data_dict['user_ntype']
item_ntype = data_dict['item_ntype']
# Assign user and movie IDs and use them as features (to learn an individual trainable
# embedding for each entity)
g.nodes[user_ntype].data['id'] = torch.arange(g.number_of_nodes(user_ntype))
g.nodes[item_ntype].data['id'] = torch.arange(g.number_of_nodes(item_ntype))
data_dict['graph'] = g
# Prepare torchtext dataset and vocabulary
if not len(item_texts):
data_dict['textset'] = None
else:
fields = {}
examples = []
for key, texts in item_texts.items():
fields[key] = torchtext.data.Field(include_lengths=True, lower=True, batch_first=True)
for i in range(g.number_of_nodes(item_ntype)):
example = torchtext.data.Example.fromlist(
[item_texts[key][i] for key in item_texts.keys()],
[(key, fields[key]) for key in item_texts.keys()])
examples.append(example)
textset = torchtext.data.Dataset(examples, fields)
for key, field in fields.items():
field.build_vocab(getattr(textset, key))
#field.build_vocab(getattr(textset, key), vectors='fasttext.simple.300d')
data_dict['textset'] = textset
return data_dict
def load_model(data_dict, device, args):
gnn = PinSAGEModel(data_dict['graph'], data_dict['item_ntype'], data_dict['textset'], args.hidden_dims, args.num_layers).to(device)
opt = torch.optim.Adam(gnn.parameters(), lr=args.lr)
checkpoint = torch.load(args.trained_path, map_location=device)
gnn.load_state_dict(checkpoint['model_state_dict'])
opt.load_state_dict(checkpoint['optimizer_state_dict'])
return gnn, opt, checkpoint['epoch']
def train(data_dict, args):
device = torch.device(args.device)
# Dataset
data_dict = prepare_dataset(data_dict, args)
g = data_dict['graph']
user_ntype = data_dict['user_ntype']
item_ntype = data_dict['item_ntype']
textset = data_dict['textset']
# Sampler
batch_sampler = sampler_module.ItemToItemBatchSampler(
g, user_ntype, item_ntype, args.batch_size)
neighbor_sampler = sampler_module.NeighborSampler(
g, user_ntype, item_ntype, args.random_walk_length,
args.random_walk_restart_prob, args.num_random_walks, args.num_neighbors,
args.num_layers)
collator = sampler_module.PinSAGECollator(neighbor_sampler, g, item_ntype, textset)
dataloader = DataLoader(
batch_sampler,
batch_size=args.batch_size,
collate_fn=collator.collate_train,
num_workers=args.num_workers)
dataloader_test = DataLoader(
torch.arange(g.number_of_nodes(item_ntype)),
batch_size=args.batch_size,
collate_fn=collator.collate_test,
num_workers=args.num_workers)
dataloader_it = iter(dataloader)
# Model
print('Loading pretrained model...')
gnn, opt, start_epoch = load_model(data_dict, device, args)
start_epoch = 0
if args.eval_epochs:
g = data_dict['graph']
item_ntype = data_dict['item_ntype']
user_ntype = data_dict['user_ntype']
user_to_item_etype = data_dict['user_to_item_etype']
timestamp = data_dict['timestamp']
nid_uid_dict = {v: k for v, k in enumerate(list(g.ndata['userID'].values())[0].numpy())}
nid_wid_dict = {nid.item(): wid.item() for wid, nid in zip(g.ndata['item_id']['item'], g.ndata['id']['item'])}
print("------Let's start Training!------")
best_recall = 0
best_hitrate = 0
for epoch in range(start_epoch, args.num_epochs + start_epoch):
gnn.train()
print('\033[95m' + f'----Epoch {epoch} starting!----' + '\033[0m')
for batch in tqdm(range(args.batches_per_epoch)):
pos_graph, neg_graph, blocks = next(dataloader_it)
for i in range(len(blocks)):
blocks[i] = blocks[i].to(device)
pos_graph = pos_graph.to(device)
neg_graph = neg_graph.to(device)
loss = gnn(pos_graph, neg_graph, blocks).to(device).mean()
opt.zero_grad()
loss.backward()
opt.step()
# print status
#if batch % 1000 == 0:
# print("num_epochs:", epoch, "||", "batches_per_epoch:", batch, "||", "loss:", loss.item())
# Evaluate
if not epoch:
continue
gnn.eval()
with torch.no_grad():
if args.eval_epochs and not epoch % args.eval_epochs:
h_item = evaluation.get_all_emb(gnn, g.ndata['id'][item_ntype],
data_dict['textset'], item_ntype, neighbor_sampler, args.batch_size, device)
item_batch = evaluation.item_by_user_batch(g, user_ntype, item_ntype, user_to_item_etype, timestamp, args)
print('\033[93m' + f'----Embedding Creation Successful----' + '\033[0m')
recalls = 0#[]
hitrates = 0
users = []
num_labels = 0
counts_n = 0
model = NearestNeighbors(n_neighbors = args.k,
metric = 'cosine',
)#cosine
model.fit(h_item.detach().cpu().numpy())
print('\033[93m' + f'----KNN fitting successful----' + '\033[0m')
for i, nodes in tqdm(enumerate(item_batch)):
# 실제 유저 ID 탐색
category = nid_uid_dict[i]
user_id = data_dict['user_category'][category] # 실제 유저 id
label = data_dict['validset'][user_id] # 테스트 라벨
users.append(user_id)
# 실제 와인 ID 탐색
item = evaluation.node_to_item(nodes, nid_wid_dict, data_dict['item_category']) # 와인 ID
label_idx = [i for i, x in enumerate(item) if x in label] # 라벨 인덱스
# 아이템 추천
nodes = [x for i, x in enumerate(nodes)] #if i not in label_idx # 라벨 인덱스 미포함 입력 학습용 노드
h_nodes = h_item[nodes]
if h_nodes.tolist() == []:
continue
h_center = torch.mean(h_nodes, axis=0) # 중앙 임베딩
_, topk = model.kneighbors(h_center.detach().cpu().numpy().reshape(1, -1))
topk = topk[0]
tp = [x for x in label if x in topk]
if not tp:
hitrates += 0
counts_n += 1
recalls += 0
else:
recalls += len(tp)
num_labels += len(label)
hitrates += 1
counts_n += 1
hitrate = hitrates / counts_n
recall = recalls / num_labels
print('\033[96m' + f'\tEpoch:{epoch}\tRecall:{recall}' + '\033[0m')
if recall > best_recall:
if not epoch % args.save_epochs:
print("This is the current best score. Save the weight.")
torch.save({'epoch': epoch,
'model_state_dict': gnn.state_dict(),
'optimizer_state_dict': opt.state_dict(),
'loss': loss},
args.save_path + '_' + str(epoch) + 'epoch.pt')
best_recall = recall
if __name__ == '__main__':
# Arguments
parser = argparse.ArgumentParser()
parser.add_argument('-d', '--dataset-path', type=str, default="graph_data/kdata_8.pkl")
parser.add_argument('--trained-path', type=str, default='model/model_78epoch.pt')
parser.add_argument('-s', '--save-path', type=str, default='model/model')
parser.add_argument('--random-walk-length', type=int, default=2)
parser.add_argument('--random-walk-restart-prob', type=float, default=0.5)
parser.add_argument('--num-random-walks', type=int, default=10)
parser.add_argument('--num-neighbors', type=int, default=5)
parser.add_argument('--num-layers', type=int, default=2)
parser.add_argument('--hidden-dims', type=int, default=1024) # 128
parser.add_argument('--batch-size', type=int, default=64)
parser.add_argument('--device', type=str, default='cuda:0') # 'cpu' or 'cuda:N'
parser.add_argument('--num-epochs', type=int, default=100)
parser.add_argument('--batches-per-epoch', type=int, default=10000)
parser.add_argument('--num-workers', type=int, default=0)
parser.add_argument('--lr', type=float, default=3e-5)
parser.add_argument('--eval-epochs', type=int, default=1)
parser.add_argument('--save-epochs', type=int, default=1)
parser.add_argument('-k', type=int, default=10)
args = parser.parse_args()
# Load dataset
with open(args.dataset_path, 'rb') as f:
dataset = pickle.load(f)
data_dict = {
'graph': dataset['train-graph'],
'val_matrix': None,
'test_matrix': None,
'item_texts': dataset['item-texts'],
'validset': dataset['validset'],
'testset': dataset['testset'],
'user_ntype': dataset['user-type'],
'item_ntype': dataset['item-type'],
'user_to_item_etype': dataset['user-to-item-type'],
'timestamp': dataset['timestamp-edge-column'],
'user_category': dataset['user-category'],
'item_category': dataset['item-category']
}
# Training
train(data_dict, args)
# torch.save({'epoch': epoch,
# 'model_state_dict': gnn.state_dict(),
# 'optimizer_state_dict': opt.state_dict(),
# 'loss': loss},
# args.save_path + '_' + str(epoch) + 'epoch.pt')