forked from bigscity-buaa/NASR
-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathtrain.py
235 lines (210 loc) · 10.2 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
import tensorflow as tf
import numpy as np
import networkx as nx
import random
import pickle
import copy
from collections import deque
import os
import math
from model import *
from utils import *
os.environ['CUDA_VISIBLE_DEVICES']='3'
# Hyper Parameters for DAN
PRE_TRAIN = False
TEST = True
RESTORE = True
GAMMA = 1.0 # discount factor for target Q
INITIAL_EPSILON = 0.5 # starting value of epsilon
FINAL_EPSILON = 0.01 # final value of epsilon
batch_size = None # size of minibatch
input_steps = None
user_size = 128
user_num = 15000
block_num = 16500
lstm_size = 384
num_layers = 2
TRAIN_BATCH_SIZE = 100 #训练输入的batch 大小
INFERENCE_BATCH_SIZE = 1 #推断的时候输入的batch 大小
PRE_EPISODE = 600
NEG_SAMPLES = 9
NEXT_ACTION_NUM = 3
train_set_size = 5000
EPISODE = 100 # Episode limitation
PRE_EPISODE = 300
TRAIN_BATCHES = 300 # Step limitation in an episode
anchor_num = 64
def train_heuristics_network(self):
self.time_step += 1
# Step 1: obtain random minibatch from replay memory
minibatch = random.sample(self.replay_buffer,BATCH_SIZE)
state_batch = [data[0] for data in minibatch]
action_batch = [data[1] for data in minibatch]
reward_batch = [data[2] for data in minibatch]
next_state_batch = [data[3] for data in minibatch]
# Step 2: calculate y
y_batch = []
Q_value_batch = self.Q_value.eval(feed_dict={self.state_input:next_state_batch})
for i in range(0,BATCH_SIZE):
done = minibatch[i][4]
if done:
y_batch.append(reward_batch[i])
else:
y_batch.append(reward_batch[i] + GAMMA * np.max(Q_value_batch[i]))
self.optimizer.run(feed_dict={
self.y_input:y_batch,
self.action_input:action_batch,
self.state_input:state_batch
})
def train_st_network(model, PRE_EPISODE):
trainData = pickle.load(open("/data/wuning/NTLR/beijing/train_loc_set", "rb"))
trainTimeData = pickle.load(open("/data/wuning/NTLR/beijing/train_time_set", "rb"))
trainUserData = pickle.load(open("/data/wuning/NTLR/beijing/train_user_set", "rb"))
trainGeoData = pickle.load(open("/data/wuning/NTLR/beijing/train_user_set", "rb"))
historyData = pickle.load(open("/data/wuning/NTLR/beijing/userIndexedHistoryAttention", "rb"))
for episode in range(PRE_EPISODE):
counter = 0
for tra_bat, time_bat, user_bat, geo_bat, his_bat, his_time_bat, his_user_bat, his_mask_bat in generate_st_batch(historyData, trainData[:train_set_size], trainTimeData[:train_set_size], trainUserData[:train_set_size], trainGeoData[:train_set_size]):
if(len(tra_bat[0]) < 15 or len(tra_bat[0]) > 100):
continue
# print(tra_bat.shape, tra_mask_bat.shape, hour_bat.shape, day_bat.shape, his_bat.shape, his_hour_bat.shape, his_day_bat.shape, his_mask_bat.shape)
_, eval_st_loss = model.session.run([model.st_all_optimizer, model.st_all_cost],feed_dict={
model.st_known_:tra_bat[:, :-1],
model.st_destination_:tra_bat[:, -1][:, np.newaxis],
model.st_output_:tra_bat[:, 1:],
model.st_time:time_bat[:, :-1],
model.st_user:user_bat[:, np.newaxis],
model.his_tra:his_bat,
model.his_time:his_time_bat,
model.his_user:his_user_bat,
model.his_padding_mask:his_mask_bat
})
# eval_st_loss = model.st_cost.eval(feed_dict={
# model.st_known_:batch[0],
# model.st_destination_:batch[2],
# model.st_output_:batch[1],
# model.trans_mat:batch[3]
# })
if counter % 100 == 0:
print("epoch:{}...".format(episode),
"batch:{}...".format(counter),
"loss:{:.4f}...".format(eval_st_loss))
counter += 1
model.all_saver.save(model.session, "/data/wuning/learnAstar/beijingComplete/pre_train_ut_attn_neural_network_epoch{}.ckpt".format(episode))
def Time_diff_attn(model):
trainData = pickle.load(open("/data/wuning/NTLR/beijing/train_loc_set", "rb"))
trainGeoData = pickle.load(open("/data/wuning/NTLR/beijing/train_geo_set", "rb"))
trainTimeData = pickle.load(open("/data/wuning/NTLR/beijing/train_time_set", "rb"))
trainUserData = pickle.load(open("/data/wuning/NTLR/beijing/train_user_set", "rb"))
historyData = pickle.load(open("/data/wuning/NTLR/beijing/userIndexedHistoryAttention", "rb"))
sub_graph_nodes = pickle.load(open("/data/wuning/NTLR/beijing/sub_graph_nodes", "rb"))
sub_graph_adjs = pickle.load(open("/data/wuning/NTLR/beijing/sub_graph_adjs", "rb"))
sub_graph_nodes = np.array(sub_graph_nodes)
sub_graph_adjs = np.array(sub_graph_adjs)
model.st_saver.restore(model.session, "/data/wuning/learnAstar/beijingComplete/pre_train_ut_attn_neural_network_epoch20.ckpt")
graphData = pickle.load(open("/data/wuning/map-matching/allGraph", "rb"))
#/data/wuning/NTLR/beijing/anchor_dbscan_set_64 /data/wuning/NASR-TKDE/data/beijing_random_anchor_set
anchor_set = pickle.load(open("/data/wuning/NTLR/beijing/anchor_dbscan_set_64", "rb"))
distMat = pickle.load(open("/data/wuning/NTLR/beijing/distMat", "rb"))
# variable_names = [v.name for v in tf.trainable_variables()]
# print(variable_names)
# anchor_set = np.array([])
location_embeddings = model.session.run(tf.get_default_graph().get_tensor_by_name("st_network/location_embedding/embeddings:0"))
location_embeddings = np.array(location_embeddings)
# anchor_set = [random.randint(0, 16499) for i in range(anchor_num)]
print("zzzzzzzzzzzzzzzzzz")
print("???", np.array(location_embeddings).shape)
adj = np.matrix(graphData)[:block_num, :block_num]
print(adj.shape)
G = nx.from_numpy_matrix(adj, create_using=nx.DiGraph())
inv_G = nx.from_numpy_matrix(adj.T, create_using=nx.DiGraph())
train_set_size = 5000
losses = []
for episode in range(PRE_EPISODE):
counter = 0
for tra_bat, time_bat, user_bat, geo_bat, his_bat, his_time_bat, his_user_bat, his_mask_bat in generate_st_batch(historyData, trainData[:train_set_size], trainTimeData[:train_set_size], trainUserData[:train_set_size], trainGeoData[:train_set_size]):
heuristics_batches = []
if(len(tra_bat[0]) < 15 or len(tra_bat[0]) > 100):
continue
_, eval_st_all_prob = model.session.run([model.st_all_optimizer, model.st_all_prob], feed_dict={
model.st_known_:tra_bat[:, :-1],
model.st_output_:tra_bat[:, 1:],
model.st_destination_:tra_bat[:, -1][:, np.newaxis],
model.st_time:time_bat[:, :-1],
model.st_user:user_bat[:, np.newaxis],
model.his_tra:his_bat,
model.his_time:his_time_bat,
model.his_user:his_user_bat,
model.his_padding_mask:his_mask_bat
})
idx = tra_bat[:, 1:]
m_idx, n_idx = idx.shape
I, J = np.ogrid[:m_idx, :n_idx]
eval_st_all_prob = eval_st_all_prob[I, J, idx]
eval_st_all_prob[eval_st_all_prob < 0.05] = 0.05
for k in range(len(tra_bat[0]) - 6, 0, -1):
item_heu_batch = - np.log(eval_st_all_prob[:, k])
for l in range(k, k + 5):
item_heu_batch += - (0.96)**(l - k) * np.log(eval_st_all_prob[:, l])
item_heu_batch = np.array(item_heu_batch)
# eval_policy[0][:, :, ]
item_known_batch = tra_bat[:, : k + 1]
item_des_batch = tra_bat[:, -1]
item_time_batch = time_bat[:, : k + 1]
item_user_batch = user_bat
item_des_nodes_batch = sub_graph_nodes[item_des_batch]
item_des_adj_batch = sub_graph_adjs[item_des_batch]
item_src_nodes_batch = sub_graph_nodes[tra_bat[:, k]]
item_src_adj_batch = sub_graph_adjs[tra_bat[:, k]]
item_heu_future_batch = tra_bat[:,: k + 6]
item_time_future_batch = time_bat[:, : k + 6]
heuristics_batches.append([item_known_batch, item_des_batch, item_heu_batch, item_time_batch, item_user_batch, item_time_future_batch, item_heu_future_batch, item_des_nodes_batch, item_src_nodes_batch, item_des_adj_batch, item_src_adj_batch])
feed_data = {}
for heu_batch in heuristics_batches:
heuristics = model.heuristics.eval(feed_dict={
model.st_known_:heu_batch[-5],
model.st_destination_:np.array(heu_batch[1])[:, np.newaxis],
model.heuristics_input:heu_batch[2],
model.st_time:heu_batch[-6],
model.st_user:heu_batch[4][:, np.newaxis],
model.his_tra:his_bat,
model.his_time:his_time_bat,
model.his_user:his_user_bat,
model.his_padding_mask:his_mask_bat,
model.src_bias_mat:heu_batch[-1],
model.des_bias_mat:heu_batch[-2],
model.src_embedding:location_embeddings[heu_batch[-3]],
model.des_embedding:location_embeddings[heu_batch[-4]],
})
#print(np.array(heu_batch[2]).shape, np.array(heuristics).shape)
heu_batch[2] += 0.82 * heuristics[:, 0]
model.optimizer.run(feed_dict = {
model.st_known_:heu_batch[0],
model.st_destination_:np.array(heu_batch[1])[:, np.newaxis],
model.heuristics_input:heu_batch[2],
model.st_time:heu_batch[3],
model.st_user:heu_batch[4][:, np.newaxis],
model.his_tra:his_bat,
model.his_time:his_time_bat,
model.his_user:his_user_bat,
model.his_padding_mask:his_mask_bat,
model.src_bias_mat:heu_batch[-1],
model.des_bias_mat:heu_batch[-2],
model.src_embedding:location_embeddings[heu_batch[-3]],
model.des_embedding:location_embeddings[heu_batch[-4]]
}
)
heuristics_batches = []
losses.append(0)
counter += 1
if counter % 100 == 0:
print("counter:", counter)
print(losses)
print("heuristics:", heuristics)
model.all_saver.save(model.session, "/data/wuning/AstarRNN/train_pgnn_geo_length_dbscan_epoch{}.ckpt".format( episode))
def main():
AstarRNN = DAN()
# train_st_network(AstarRNN, 50)
Time_diff_attn(AstarRNN)
if __name__ == '__main__':
main()