/
helper.py
199 lines (158 loc) · 8.96 KB
/
helper.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
# !/usr/bin/env python
# -*- coding:utf-8 -*-
import torch
import torch.nn as nn
import numpy as np
import torch.optim as optim
from tools.metrics import masked_mae_torch, masked_mape_torch, masked_rmse_torch, metric, metric_all
from tools.utils import StepLR2
class Trainer():
def __init__(self,
CausalHMM,
base_lr,
weight_decay,
milestones,
lr_decay_ratio,
min_learning_rate,
max_grad_norm,
num_for_target,
num_for_predict,
scaler,
device,
loss_weight,
DAG_loss_weight=5,
):
self.scaler = scaler
self.CausalHMM = CausalHMM
self.device = device
self.max_grad_norm = max_grad_norm
self.loss_weight = loss_weight
self.DAG_loss_weight = DAG_loss_weight
self.CausalHMM.to(device)
self.CausalHMM_optimizer = optim.Adam(self.CausalHMM.parameters(), lr=base_lr, weight_decay=weight_decay)
self.CausalHMM_scheduler = StepLR2(optimizer=self.CausalHMM_optimizer,
milestones=milestones,
gamma=lr_decay_ratio,
min_lr=min_learning_rate)
self.SmoothL1loss = nn.SmoothL1Loss(reduction='mean')
self.scaler = scaler
self.num_for_target = num_for_target
self.num_for_predict = num_for_predict
def train(self, tpos, weather, bike_flow, taxi_flow, bus_flow, speed):
"""
:param input: [batch node time hdim]
output: [batch node time hdim]
:param real_val:
:return:
"""
self.CausalHMM.train()
self.CausalHMM_optimizer.zero_grad()
po_z_mu_list, po_z_var_list, po_z_list, \
eps_kl_loss, z_kl_loss, \
rec_all, gen_all = self.CausalHMM(tpos, weather,
self.scaler[0].transform(bike_flow),
self.scaler[1].transform(taxi_flow),
self.scaler[2].transform(bus_flow),
self.scaler[3].transform(speed),
test=False)
rec_bike = self.scaler[0].inverse_transform(rec_all[0])
rec_taxi = self.scaler[1].inverse_transform(rec_all[1])
rec_bus = self.scaler[2].inverse_transform(rec_all[2])
rec_speed = self.scaler[3].inverse_transform(rec_all[3])
rec_loss = 0
rec_loss += self.SmoothL1loss(rec_bike, bike_flow[:, :, :self.num_for_predict, :])
rec_loss += self.SmoothL1loss(rec_taxi, taxi_flow[:, :, :self.num_for_predict, :])
rec_loss += self.SmoothL1loss(rec_bus, bus_flow[:, :, :self.num_for_predict, :])
rec_loss += self.SmoothL1loss(rec_speed, speed[:, :, :self.num_for_predict, :])
rec_mae, rec_rmse, rec_mape = metric_all([rec_bike, rec_taxi, rec_bus, rec_speed],
[bike_flow[:, :, :self.num_for_predict, :],
taxi_flow[:, :, :self.num_for_predict, :],
bus_flow[:, :, :self.num_for_predict, :],
speed[:, :, :self.num_for_predict, :]])
# rec_loss /= 4.
pred_loss = 0
gen_mae = {}
gen_rmse = {}
gen_mape = {}
gen_bike = self.scaler[0].inverse_transform(gen_all[0])
gen_taxi = self.scaler[1].inverse_transform(gen_all[1])
gen_bus = self.scaler[2].inverse_transform(gen_all[2])
gen_speed = self.scaler[3].inverse_transform(gen_all[3])
pred_loss += self.SmoothL1loss(gen_bike, bike_flow[:, :, 1:self.num_for_predict + 1, :])
pred_loss += self.SmoothL1loss(gen_taxi, taxi_flow[:, :, 1:self.num_for_predict + 1, :])
pred_loss += self.SmoothL1loss(gen_bus, bus_flow[:, :, 1:self.num_for_predict + 1, :])
pred_loss += self.SmoothL1loss(gen_speed, speed[:, :, 1:self.num_for_predict + 1, :])
gen_mae, gen_rmse, gen_mape = metric_all([gen_bike, gen_taxi, gen_bus, gen_speed],
[bike_flow[:, :, 1:self.num_for_predict + 1, :],
taxi_flow[:, :, 1:self.num_for_predict + 1, :],
bus_flow[:, :, 1:self.num_for_predict + 1, :],
speed[:, :, 1:self.num_for_predict + 1, :]])
HMM_loss = eps_kl_loss + z_kl_loss + rec_loss
HMM_loss = HMM_loss + self.DAG_loss_weight * self.CausalHMM.SCM.cal_loss()
total_loss = HMM_loss + self.loss_weight * pred_loss
total_loss.backward()
if self.max_grad_norm is not None:
torch.nn.utils.clip_grad_norm_(self.CausalHMM.parameters(), self.max_grad_norm)
self.CausalHMM_optimizer.step()
return total_loss.item(), eps_kl_loss.item(), z_kl_loss.item(), rec_loss.item(), pred_loss.item(), \
rec_mae, rec_rmse, rec_mape, \
gen_mae, gen_rmse, gen_mape
def eval(self, tpos, weather, bike_flow, taxi_flow, bus_flow, speed):
'''
:param tpos:
:param weather:
:param bike_flow:
:param taxi_flow:
:param bus_flow:
:param speed:
:return:
'''
self.CausalHMM.eval()
with torch.no_grad():
po_z_mu_list, po_z_var_list, po_z_list, \
eps_kl_loss, z_kl_loss, \
rec_all, gen_all = self.CausalHMM(tpos, weather,
self.scaler[0].transform(bike_flow),
self.scaler[1].transform(taxi_flow),
self.scaler[2].transform(bus_flow),
self.scaler[3].transform(speed),
test=True)
rec_bike = self.scaler[0].inverse_transform(rec_all[0])
rec_taxi = self.scaler[1].inverse_transform(rec_all[1])
rec_bus = self.scaler[2].inverse_transform(rec_all[2])
rec_speed = self.scaler[3].inverse_transform(rec_all[3])
rec_loss = 0
rec_loss += self.SmoothL1loss(rec_bike, bike_flow[:, :, :self.num_for_predict, :])
rec_loss += self.SmoothL1loss(rec_taxi, taxi_flow[:, :, :self.num_for_predict, :])
rec_loss += self.SmoothL1loss(rec_bus, bus_flow[:, :, :self.num_for_predict, :])
rec_loss += self.SmoothL1loss(rec_speed, speed[:, :, :self.num_for_predict, :])
rec_mae, rec_rmse, rec_mape = metric_all([rec_bike, rec_taxi, rec_bus, rec_speed],
[bike_flow[:, :, :self.num_for_predict, :],
taxi_flow[:, :, :self.num_for_predict, :],
bus_flow[:, :, :self.num_for_predict, :],
speed[:, :, :self.num_for_predict, :]])
pred_loss = 0
gen_mae = {}
gen_rmse = {}
gen_mape = {}
pred = torch.zeros_like(speed[:, :, 1:self.num_for_predict+1, :])
gen_bike = self.scaler[0].inverse_transform(gen_all[0])
gen_taxi = self.scaler[1].inverse_transform(gen_all[1])
gen_bus = self.scaler[2].inverse_transform(gen_all[2])
gen_speed = self.scaler[3].inverse_transform(gen_all[3])
pred = torch.cat([gen_bike, gen_taxi, gen_bus, gen_speed], dim=-1)
pred_loss += self.SmoothL1loss(gen_bike, bike_flow[:, :, 1:self.num_for_predict + 1, :])
pred_loss += self.SmoothL1loss(gen_taxi, taxi_flow[:, :, 1:self.num_for_predict + 1, :])
pred_loss += self.SmoothL1loss(gen_bus, bus_flow[:, :, 1:self.num_for_predict + 1, :])
pred_loss += self.SmoothL1loss(gen_speed, speed[:, :, 1:self.num_for_predict + 1, :])
gen_mae, gen_rmse, gen_mape = metric_all([gen_bike, gen_taxi, gen_bus, gen_speed],
[bike_flow[:, :, 1:self.num_for_predict + 1, :],
taxi_flow[:, :, 1:self.num_for_predict + 1, :],
bus_flow[:, :, 1:self.num_for_predict + 1, :],
speed[:, :, 1:self.num_for_predict + 1, :]])
HMM_loss = eps_kl_loss + z_kl_loss + rec_loss
HMM_loss = HMM_loss + self.DAG_loss_weight * self.CausalHMM.SCM.cal_loss()
total_loss = HMM_loss + self.loss_weight * pred_loss
return total_loss.item(), eps_kl_loss.item(), z_kl_loss.item(), rec_loss.item(), pred_loss.item(), \
rec_mae, rec_rmse, rec_mape, \
gen_mae, gen_rmse, gen_mape, pred