张量张量张量值未分配

发布于 2025-01-31 06:37:41 字数 693 浏览 2 评论 0原文

我正在使用以下自定义损失函数：
model.compile（lose = custom_loss，...）
自定义损失函数定义为：

def custom_loss(y_actual,y_pred):
    custom_loss = loss_function(y_actual,y_pred) + penalty * quantizer.model_size()
    return custom_loss

其中Qualtizer是类的对象，model_size（）是此类的一种方法。在此model_size（）方法中，我正在尝试创建一些具有一些初始值的张量变量。但是似乎没有分配值。我正在使用打印（tf.constant（1.0，dtype = float））） and print（tf.variable（1.0，dtype = float））查看张量输出就像张量（“ custom_loss/const：0”，shape =（），dtype = float32） and code>＆lt; tf.variable'custic_loss/varible/variable/varible：0'shape =（）dtype = float32＆gt; 。是什么原因导致这个问题？

原文

I'm using a custom loss function as below:
model.compile(loss=custom_loss,...)

And the custom loss function is defined as:

def custom_loss(y_actual,y_pred):
    custom_loss = loss_function(y_actual,y_pred) + penalty * quantizer.model_size()
    return custom_loss

where quantizer is a object of a class and model_size() is a method of this class. In this model_size() method, I'm trying to create some tensors variable with some initial value. But it seems like the value is not assigned. I'm printing using print(tf.constant(1.0, dtype=float)) and print(tf.Variable(1.0, dtype=float)) to see the tensors and the output is like Tensor("custom_loss/Const:0", shape=(), dtype=float32) and <tf.Variable 'custom_loss/Variable:0' shape=() dtype=float32>. What causes this issue?