-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathutil.py
196 lines (182 loc) · 7.78 KB
/
util.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
import copy
from itertools import chain
import pandas as pd
import numpy as np
import torch
from matplotlib import pyplot as plt
from scipy.interpolate import make_interp_spline
from torch import nn
from torch.optim.lr_scheduler import StepLR
from tqdm import tqdm
import time
from models import LSTM, BiLSTM, Seq2Seq, RMSELoss, MAE
from preprocess.feature_engineering import Input_Module
device = torch.device("cuda:0") if torch.cuda.is_available() else torch.device("cpu")
def get_val_loss(args, model, Val):
label = []
val_loss = []
# 将模型设置为评估模式
model.eval()
loss_function = RMSELoss().to(device)
print('validating...')
val_total_time = 0
input_module = Input_Module()
for (seq, target) in tqdm(Val):
currt_time = time.time()
seq, target, type_data = input_module(seq, target)
seq = seq.to(device)
target = target.to(device)
type_data = type_data.to(device)
with torch.no_grad():
y_pred = model(seq, target, type_data, False)
loss = loss_function(y_pred, target)
val_loss.append(loss.item())
next_time = time.time()
use_time = next_time - currt_time
val_total_time += use_time
return np.mean(val_loss), val_total_time
# 计算MAPE指标的函数
def get_mape(y_true, y_pred):
"""
Compute mean absolute percentage error (MAPE)
"""
y_true, y_pred = np.array(y_true), np.array(y_pred)
return np.mean(np.abs((y_true - y_pred) / y_true)) * 100
def load_model(args):
input_size, hidden_size, num_layers = args.input_size, args.hidden_size, args.num_layers
output_size = args.output_size
if args.flag in ['us', 'ms', 'mm']:
if args.bidirectional:
model = BiLSTM(input_size, hidden_size, num_layers, output_size, batch_size=args.batch_size).to(device)
else:
model = LSTM(input_size, hidden_size, num_layers, output_size, batch_size=args.batch_size).to(device)
elif args.flag in ['seq2seq']:
model = Seq2Seq(input_size, hidden_size, num_layers, output_size, batch_size=args.batch_size).to(device)
else:
model = LSTM(input_size, hidden_size, num_layers, output_size, batch_size=args.batch_size).to(device)
return model
def train(args, Dtr, Val, path):
model = load_model(args)
loss_function = RMSELoss().to(device)
# optimizer是优化器，用来更新模型参数，作用是最小化损失函数
if args.optimizer == 'adam':
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr,
weight_decay=args.weight_decay)
else:
optimizer = torch.optim.SGD(model.parameters(), lr=args.lr,
momentum=0.9, weight_decay=args.weight_decay)
# scheduler为学习率调度器 step_size是学习率衰减的步数 gamma是学习率衰减的比例
scheduler = StepLR(optimizer, step_size=args.step_size, gamma=args.gamma)
# training
min_epochs = -1
best_model = None
min_val_loss = 5
best_loss = 5
best_epoch = 0
train_total_time = 0
val_total_time = 0
for epoch in tqdm(range(args.epochs)):
train_loss = []
input_module = Input_Module()
# seq:(128,10,13) label:(128,5,13)
for step, (seq, label) in enumerate(Dtr):
# 返回当前时间的时间戳
current_time = time.time()
# print('seq:', seq.size())
# print('label:', label.size())
# print()
# 输出后seq:(128,10,5) label:(128,5,2) type_data:(128,10,14)
seq, label, type_data = input_module(seq, label) #TODO: input_module部分的三个输出有疑问
# print('---------------------------------------')
# print()
# print('seq:', seq.size())
# print('label:', label.size())
# print('type_data:', type_data.size())
seq = seq.to(device)
label = label.to(device)
type_data = type_data.to(device)
y_pred = model(seq, label, type_data)
# 求loss
loss = loss_function(y_pred, label)
# loss.item()用来取loss的值，精度比用索引来的高，在求损失函数等时我们一般用.item()
train_loss.append(loss.item())
# 梯度初始化为零
optimizer.zero_grad()
# 反向传播求梯度
loss.backward()
# 更新所有参数
optimizer.step()
next_time = time.time()
ues_time = next_time - current_time
train_total_time += ues_time
step += 1
if step % 100 == 0:
print("epoch :{},step :{} ,Train loss: {}".format(epoch, step, np.sum(train_loss) / step))
# 更新学习率
scheduler.step()
# validation
val_loss, val_time = get_val_loss(args, model, Val)
if epoch > min_epochs and val_loss < min_val_loss:
min_val_loss = val_loss
best_model = copy.deepcopy(model)
val_total_time += val_time
print('epoch {:03d} train_loss {:.8f} val_loss {:.8f}'.format(epoch, np.mean(train_loss), val_loss))
model.train()
state = {'models': best_model.state_dict()}
print('best_model:{}, min_val_loss:{}, train time:{}, val time:{}'.format(best_model, min_val_loss,
train_total_time / args.epochs,
val_total_time / args.epochs))
# 保存学习到的参数
torch.save(state, path)
def test(args, Dte, path, m, n):
print('输入任意字符开始测试：')
a = input()
# y = []
test_loss = []
mae_loss = []
loss_t_dict = {}
loss_dict = {}
print('loading models...')
loss_function = RMSELoss().to(device)
mae_loss_function = MAE().to(device)
model = load_model(args)
model.load_state_dict(torch.load(path, map_location='cpu')['models'])
model.eval()
test_total_time = 0
print('predicting...')
test_total_time = 0
input_module = Input_Module()
for (seq, target) in tqdm(Dte):
current_time = time.time()
seq, target, type_data = input_module(seq, target)
# target = list(chain.from_iterable(target.data.tolist()))
type_data = type_data.to(device)
# y.extend(target)
seq = seq.to(device)
target = target.to(device)
seq_len = target.shape[1]
with torch.no_grad():
# print(seq.shape)
# print(target.shape)
# print(type_data.shape)
y_pred = model(seq, target, type_data, False)
for t in range(seq_len):
if t not in loss_dict.keys():
loss_dict[t] = []
loss_dict[t].append(loss_function(y_pred[:, t, :], target[:, t, :]).item())
if t not in loss_t_dict.keys():
loss_t_dict[t] = []
loss_t_dict[t].append(loss_function(y_pred[:, :t + 1, :], target[:, :t + 1, :]).item())
loss = loss_function(y_pred, target)
test_loss.append(loss.item())
mae = mae_loss_function(y_pred, target)
mae_loss.append(mae.item())
next_time = time.time()
ues_time = next_time - current_time
test_total_time += ues_time
for i, loss_ in loss_dict.items():
print('predict the {} point mean loss is {}'.format(i + 1, np.mean(loss_)))
for i, _loss in loss_t_dict.items():
print('test predict the first {} points` mean loss is {}'.format(i + 1, np.mean(_loss)))
print('test mean rmse is {}, mae is {}, test time is {}'.format(np.mean(test_loss), np.mean(mae_loss),
test_total_time))