如何为多维目标定义自定义损失函数

时间:2020-04-28 13:20:23

标签: deep-learning neural-network tensorflow2.0 loss-function tf.keras

我正在使用Tensorflow 2.0,并且使用的是常规顺序层。

我正在尝试定义一个自定义损失函数,该函数将执行以下操作:

  • 采用输入的某些元素
  • 计算它们的总和并反转结果
  • 将结果与y_pred的一部分相乘
  • 将结果约束为尽可能接近1

因此损失函数为L()= MSE()+(如上所述)

我的代码如下:

# Creates the folder structure defined in folder structure section below
function createFolderStructure() {
     depth="1"
     while (( "$#" )); do
         while (( $1 != $depth )); do
             cd ..
             (( depth-- ))
         done
         shift
         mkdir "$1"
         cd "$1"
         (( depth++ ))
         shift
         shift
         out=""
         while [[ "$1" != "-" ]]; do
             out=$out" ""$1"
             shift
         done
         shift
         echo "$out" > README.md
     done
     while (( 1 != $depth )); do
         cd ..
         (( depth-- ))
     done
}

# If you like you can read in user defined values here and use them as variables in the folder structure section, e.g.
# echo -n "Enter month of films"
# read month
# ...
# 1 shared - Folder for shared stuff -
#    2 $month - Films from month $month - 
#       3 projects - Folder for projects -
# ... 

# Folder Structure Section
read -r -d '' FOLDERSTRUCTURE << EOM
1 shared - Folder for shared stuff -
     2 projects - Folder for projects -
          3 movies - Folder for movies -
                4 action - Folder for action movies -
     2 series - Folder for series -
     2 backup - Backup folder -
EOM

createFolderStructure $FOLDERSTRUCTURE

我想知道我在做什么是否正确。没有错误,训练进行了,但是我不知道这段代码是否符合我的定义。通常,如果考虑到成批的数据而不是单个记录,则可以正常工作

0 个答案:

没有答案
相关问题