Python - RNN LSTM模型精度低

时间:2018-05-20 12:16:08

标签: keras deep-learning lstm rnn

我尝试使用此数据集样本构建LSTM模型

  

(患者编号,以毫米/秒为单位的时间,X Y和Z的标准化,   分别为峰度,偏度,俯仰,滚动和偏航,标签。

     

1.15,-0.248010047716,0.00378335508419,-0.0152548459993,-86.3738760481,0.872322164158,-3.51314800063,0

     

1,31,-0.248010047716,0.00378335508419,-0.0152548459993,-86.3738760481,0.872322164158,-3.51314800063,0

     

1,46,-0.267422664673,0.0051143782875,-0.0191247001961,-85.7662354031,1.0928406847,-4.08015176908,0

     

1.62,-0.267422664673,0.0051143782875,-0.0191247001961,-85.7662354031,1.0928406847,-4.08015176908,0

这就是我用代码做的事情

$jalali_date = explode("/", request()->to);

$gregorian_date_time = \jDateTime::toGregorianDate($jalali_date[0], $jalali_date[1], $jalali_date[2])->setTime(23, 55)->format("Y-m-d H:i");
return $gregorian_date_time;

没有错误但准确度非常低且损失非常高

  

大纪元1/20     - 63s - 损失:15.0343 - acc:0.0570 Epoch 2/20     - 60s - 损失:15.0343 - acc:0.0570 Epoch 3/20     - 60s - 损失:15.0343 - acc:0.0570 Epoch 4/20     - 60s - 损失:15.0343 - acc:0.0570

1 个答案:

答案 0 :(得分:0)

这里的错误是使用softmax激活函数,因为它用于分类问题..但这是一个二元问题所以最好的激活函数是sigmoid