如何解决缺少值的自动编码器Keras，Python
permno 10001 10002 10012 ... 93434 93435 93436
date ...
2005-01-31 -0.040580 -0.132466 -0.320388 ... NaN NaN NaN
2005-02-28 -0.045166 -0.033255 -0.285714 ... NaN NaN NaN
2005-03-31 0.124822 -0.013081 -0.080000 ... NaN NaN NaN
2005-04-29 -0.074684 -0.066700 -0.195652 ... NaN NaN NaN
2005-05-31 0.219030 0.046586 -0.243243 ... NaN NaN NaN
... ... ... ... ... ... ...
2019-08-30 NaN NaN NaN ... -0.104730 NaN -0.066222
2019-09-30 NaN NaN NaN ... -0.101887 NaN 0.067639
2019-10-31 NaN NaN NaN ... -0.050420 NaN 0.307427
2019-11-29 NaN NaN NaN ... 0.000000 NaN 0.047695
2019-12-31 NaN NaN NaN ... -0.070796 NaN 0.267897
from keras import layers
from keras.layers import Input,Dense
from keras import regularizers,models,optimizers
import matplotlib.pyplot as plt
from datetime import datetime
import seaborn as sns
def lin_ae(object,neurons,learning_rate = 1e-4,regularization = 1e-5,epochs=1000):
input_layer = Input(shape=(object.shape[1],))
encoder = Dense(neurons,activation='linear',kernel_regularizer=regularizers.l2(regularization))(input_layer)
decoder = Dense(object.shape[1],kernel_regularizer=regularizers.l2(regularization))(encoder)
autoencoder = models.Model(input_layer,decoder)
autoencoder.compile(optimizer=optimizers.adam(lr=learning_rate),loss='mean_squared_error')
#Fit the model
autoencoder.fit(object,object,epochs=epochs,batch_size=4,shuffle=True)
return autoencoder
如您所见，我的数据帧非常不平衡。当我尝试使用lin_ae(object=df.values,neurons=5,regularization=0,epochs=1000)运行上面定义的函数时
我看到损失函数缺少值：
...
Epoch 103/1000
180/180 [==============================] - 0s 756us/step - loss: nan
Epoch 104/1000
180/180 [==============================] - 0s 756us/step - loss: nan
...
那我该怎么办？我当然可以丢弃缺少值的观察值，但是我想保持时间和permno（列）的尺寸很大。我该如何告诉keras函数一次“忽略”缺失的值，而不是为了摆脱NaN而不得不丢弃整个数据帧的行或列？
版权声明：本文内容由互联网用户自发贡献，该文观点与技术仅代表作者本人。本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如发现本站有涉嫌侵权/违法违规的内容， 请发送邮件至 dio@foxmail.com 举报，一经查实，本站将立刻删除。
