forked from gcooq/SelfTrip
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpoi_embedding.py
176 lines (142 loc) · 6.65 KB
/
poi_embedding.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
#!/usr/bin/python3
# coding=utf-8
import pandas as pd
import numpy as np
import tensorflow as tf
from tensorflow.keras import Model
from tensorflow.keras.layers import Dot, Embedding, Flatten
import os
# os.environ["CUDA_VISIBLE_DEVICES"]="-1"
poi_embedding_size = 250
class Word2Vec(Model):
def __init__(self, vocab_size, embedding_dim, num_ns):
super(Word2Vec, self).__init__()
self.query_embedding = Embedding(vocab_size,
embedding_dim,
input_length=1,
name="query_em_layer")
self.poi_embedding = Embedding(vocab_size,
embedding_dim,
input_length=num_ns+1,
name='poi_em_layer')
self.dots = Dot(axes=(3, 1))
self.flatten = Flatten()
def call(self, pair):
target, context = pair
word_emb0 = self.query_embedding(target[:, 0])
word_emb1 = self.query_embedding(target[:, 1])
word_emb = (word_emb0 + word_emb1)/2
context_emb = self.poi_embedding(context)
dots = self.dots([context_emb, word_emb])
return self.flatten(dots)
class PoiEmbedding():
def __init__(self, dis_matric, train_data, poi_num, num_ns=3, walk_num=10, walk_length=6):
self.dis_matric = dis_matric
self.walk_num = walk_num
self.walk_length = walk_length
self.train_data = train_data
self.poi_num = poi_num
self.num_ns = num_ns
self.sentences = []
self.positive_data = []
self.dataset = None
def gen_sentences(self, start, end):
for i in range(self.walk_num):
sentence = [start, ]
poi_now = start
for j in range(self.walk_length):
if np.sum(self.poi_graph.loc[poi_now]) != 1:
break
poi_next = np.random.choice(self.poi_graph.columns, p=self.poi_graph.loc[poi_now])
if poi_next == end:
sentence.append(poi_next)
break
sentence.append(poi_next)
poi_now = poi_next
if (len(sentence) >= 3 and sentence[-1] == end):
self.sentences.append(sentence)
def gen_train(self):
self.poi_graph = pd.DataFrame(np.zeros(self.poi_num**2).reshape(self.poi_num,self.poi_num),
index=range(0, self.poi_num),
columns=range(0, self.poi_num))
for traj in self.train_data:
for i in range(len(traj)-1):
if(traj[i+1] == 0):
break
self.poi_graph.loc[traj[i],traj[i+1]] += 1
self.poi_graph = self.poi_graph + self.dis_matric
for i in self.poi_graph.index:
if np.sum(self.poi_graph.loc[i]) == 0:
continue
self.poi_graph.loc[i] = self.poi_graph.loc[i] / np.sum(self.poi_graph.loc[i])
print(self.poi_graph)
for start in self.poi_graph.index:
for end in self.poi_graph.columns:
self.gen_sentences(start, end)
print('deepwalk data ', len(self.sentences))
for sentence in self.sentences:
target = (sentence[0],sentence[-1])
for poi in sentence[1:-1]:
self.positive_data.append((target,poi))
print('正样本：',len(self.positive_data))
targets, contexts, labels = [], [], []
for target_pois, context_poi in self.positive_data:
context_class = tf.expand_dims(
tf.constant([context_poi], dtype="int64"), 1)
negative_sampling_candidates, _, _ = tf.random.log_uniform_candidate_sampler(
true_classes=context_class,
num_true=1,
num_sampled=self.num_ns,
unique=True,
range_max=self.poi_num,
seed=40,
name="nagetive_sampling")
# Build context and label vectors (for one target word)
negative_sampling_candidates = tf.expand_dims(
negative_sampling_candidates, 1)
context = tf.concat([context_class, negative_sampling_candidates], 0)
label = tf.constant([1] + [0] * self.num_ns, dtype="int64")
# Append each element from the training example to global lists.
targets.append(target_pois)
contexts.append(context)
labels.append(label)
print(len(targets),len(contexts),len(labels))
BATCH_SIZE = 4
BUFFER_SIZE = 10000
self.dataset = tf.data.Dataset.from_tensor_slices(((targets, contexts), labels))
self.dataset = self.dataset.shuffle(BUFFER_SIZE).batch(BATCH_SIZE, drop_remainder=True)
def train(self, city, em_size):
embedding_dim = em_size # poi嵌入维度
word2vec = Word2Vec(self.poi_num, embedding_dim, self.num_ns)
word2vec.compile(optimizer=tf.keras.optimizers.Adam(lr=0.001),
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir="logs")
word2vec.fit(self.dataset, epochs=20, callbacks=[tensorboard_callback])
que_weights = word2vec.get_layer('query_em_layer').get_weights()[0]
poi_weights = word2vec.get_layer('poi_em_layer').get_weights()[0]
que_weights = pd.DataFrame(que_weights)
print(que_weights.shape)
poi_weights = pd.DataFrame(poi_weights)
print(poi_weights.shape)
# que_weights.to_csv('./self-embedding/'+city+'_que_weight.csv',index=False)
# poi_weights.to_csv('./self-embedding/'+city+'_poi_weight.csv',index=False)
if __name__ == '__main__':
city = 'Osak'
# city = 'Glas'
# city = 'Edin'
# city = 'Toro'
trajs_data = open('./train_data/'+city+'-trajs.dat','r')
trajs_list = []
poi_dis_matric = pd.read_csv('./dis_matric/' + city + '_dis_matric.csv')
for line in trajs_data.readlines():
tlist = [eval(i) for i in line.split()]
trajs_list.append(tlist)
print('total number', len(trajs_list))
poi_size = poi_dis_matric.shape[0] # poi个数
print('poi number', poi_size)
poi_dis_matric.index = [i for i in range(0, poi_size)]
poi_dis_matric.columns = [i for i in range(0, poi_size)]
self_embedding = PoiEmbedding(poi_dis_matric, trajs_list, poi_size)
self_embedding.gen_train()
self_embedding.train(city, poi_embedding_size)