1.网络结构
U是LSTM的一种变体，可以说是简化版本的LSTM，但是预测效果也很不错，因此更常用。
GRU使记忆体
h
t
h^{t}
ht融合了长期记忆和短期记忆。
(1)记忆体
h
t
h^{t}
ht
h
t
=
z
t
⊙
h
t
?
1
+
(
1
?
z
t
)
⊙
h
t
^
h^{t}=z^{t}\odot h^{t-1}+(1-z^{t})\odot \widehat{h^{t}}
ht=zt⊙ht?1+(1?zt)⊙ht
其中，
z
t
z^{t}
zt为更新门，控制当前状态需要从历史状态中保留多少信息，以及需要从候选状态中接受多少新信息。
(2)候选状态
h
t
^
\widehat{h^{t}}
ht
h
t
^
=
tanh
?
(
W
h
x
t
+
U
(
r
t
⊙
h
t
?
1
)
+
b
h
)
\widehat{h^{t}}=\tanh(\mathbf{W}_{h}x^{t}+\mathbf{U}(r^{t}\odot h^{t-1})+\mathbf{b}_{h})
ht
=tanh(Wh?xt+U(rt⊙ht?1)+bh?)
其中，
r
t
r^{t}
rt为
t
t
t时刻的重置门。
(3)重置门和更新门
重置门用来控制候选状态
h
t
^
\widehat{h^{t}}
ht
的计算是否依赖上一时刻的记忆体:
r
t
=
s
i
g
m
o
i
d
(
W
r
x
t
+
U
r
h
t
?
1
+
b
r
)
r^{t}=sigmoid(\mathbf{W}_{r}x^{t}+\mathbf{U}_{r}h^{t-1}+\mathbf{b}_{r})
rt=sigmoid(Wr?xt+Ur?ht?1+br?)
从计算公式可知，
r
t
∈
[
0
,
1
]
r^{t}\in [0,1]
rt∈[0,1]。当
r
t
=
0
r^{t}=0
rt=0时，候选状态和历史状态无关；当
r
t
=
1
r^{t}=1
rt=1时，候选状态和简单循环网络一致。
更新门的作用和求解方式与重置门相同，计算过程如下：
z
t
=
s
i
g
m
o
i
d
(
W
z
x
t
+
U
z
h
t
?
1
+
b
z
)
z^{t}=sigmoid(\mathbf{W}_{z}x^{t}+\mathbf{U}_{z}h^{t-1}+\mathbf{b}_{z})
zt=sigmoid(Wz?xt+Uz?ht?1+bz?)
2.前向传播过程
(1)计算更新门和重置门：
r
t
=
s
i
g
m
o
i
d
(
W
r
x
t
+
U
r
h
t
?
1
+
b
r
)
z
t
=
s
i
g
m
o
i
d
(
W
z
x
t
+
U
z
h
t
?
1
+
b
z
)
r^{t}=sigmoid(\mathbf{W}_{r}x^{t}+\mathbf{U}_{r}h^{t-1}+\mathbf{b}_{r})\\ z^{t}=sigmoid(\mathbf{W}_{z}x^{t}+\mathbf{U}_{z}h^{t-1}+\mathbf{b}_{z})
rt=sigmoid(Wr?xt+Ur?ht?1+br?)zt=sigmoid(Wz?xt+Uz?ht?1+bz?)
(2)通过重置门和上一时刻记忆体，更新候选状态：
h
t
^
=
tanh
?
(
W
h
x
t
+
U
(
r
t
⊙
h
t
?
1
)
+
b
h
)
\widehat{h^{t}}=\tanh(\mathbf{W}_{h}x^{t}+\mathbf{U}(r^{t}\odot h^{t-1})+\mathbf{b}_{h})
ht
=tanh(Wh?xt+U(rt⊙ht?1)+bh?)
(3)计算当前时刻记忆体:
h
t
=
z
t
⊙
h
t
?
1
+
(
1
?
z
t
)
⊙
h
t
^
h^{t}=z^{t}\odot h^{t-1}+(1-z^{t})\odot \widehat{h^{t}}
ht=zt⊙ht?1+(1?zt)⊙ht
3.keras+GPU茅台股票预测
#导入工具包
import pandas as pd
maotai=pd.read_csv('./SH600519.csv')
training_set = maotai.iloc[0:2126,2:3].values
test_set = maotai.iloc[2126:,2:3].values
print(training_set.shape,test_set.shape)
#数据归一化
from sklearn.preprocessing import MinMaxScaler
print(training_set.max(),training_set.min())
sc=MinMaxScaler(feature_range=(0,1))
training_set=sc.fit_transform(training_set)
test_set=sc.fit_transform(test_set)
print(training_set.max(),training_set.min())
#划分数据集测试集
#调整数据维度
import numpy as np
import tensorflow as tf
x_train,y_train,x_test,y_test=[],[],[],[]
for i in range(60,len(training_set)):
x_train.append(training_set[i-60:i,0])
y_train.append(training_set[i,0])
np.random.seed(7)
np.random.shuffle(x_train)
np.random.seed(7)
np.random.shuffle(y_train)
tf.random.set_seed(7)
x_train,y_train = np.array(x_train),np.array(y_train)
x_train = np.reshape(x_train, (x_train.shape[0], 60, 1))
for i in range(60, len(test_set)):
x_test.append(test_set[i - 60:i, 0])
y_test.append(test_set[i, 0])
x_test, y_test = np.array(x_test), np.array(y_test)
x_test = np.reshape(x_test, (x_test.shape[0], 60, 1))
#搭建GRU网络
from tensorflow.keras.layers import GRU,Dropout,Dense
model = tf.keras.Sequential([
GRU(80,return_sequences=True),
Dropout(0.2),
GRU(100),
Dropout(0.2),
Dense(1)
])
#配置网络
model.compile(optimizer=tf.keras.optimizers.Adam(0.001),
loss='mean_squared_error')
#训练网络
history = model.fit(x_train, y_train, batch_size=64, epochs=50,
validation_data=(x_test, y_test), validation_freq=1)
#绘制Loss曲线
loss = history.history['loss']
val_loss = history.history['val_loss']
plt.plot(loss,label='Training Loss')
plt.plot(val_loss,label='Validation Loss')
plt.legend()
plt.title('Loss')
plt.show()
#预测test并且和真实标签对比
predict_price = model.predict(x_test)
predict_price = sc.inverse_transform(predict_price)
real_price = sc.inverse_transform(test_set[60:])
plt.plot(real_price, color='red', label='MaoTai Stock Price')
plt.plot(predict_price, color='blue', label='Predicted MaoTai Stock Price')
plt.title('MaoTai Stock Price Prediction')
plt.xlabel('Time')
plt.ylabel('MaoTai Stock Price')
plt.legend()
plt.show()
|