自定义模型中遇到的报错日志及解决方法（TensorFlow2.1.0 ）

知识不足恐惧症

已于 2022-07-12 21:21:39 修改

阅读量1.1k

点赞数

分类专栏：错误日志解决文章标签： tensorflow 深度学习 python

于 2022-07-12 21:06:14 首次发布

本文链接：https://blog.csdn.net/yaogepila/article/details/125751034

版权

错误日志解决专栏收录该内容

3 篇文章 0 订阅

订阅专栏

1. ValueError: Layer dense expects 1 inputs, but it received 2 input tensors.

问题代码：

model.build(input_shape=[None, 50])

问题出在参数input_shape的传入上，不能传入列表形式的shape，只能是tuple类型的。
修改后正确代码为

model.build(input_shape=(None, 50))

2.ValueError: You tried to call count_params on *, but the layer isn’t built. You can build it manually via: **

问题代码：

import tensorflow as tf
from tensorflow.keras import layers, models


class MyModel(tf.keras.models.Model):
    def __init__(self):
        super(MyModel, self).__init__()

    def build(self, input_shape):
        # self.dense = layers.Dense(4)
        # self.dense2 = layers.Dense(2)
        # self.built = True
        super(MyModel, self).build(input_shape)

    def call(self, inputs):

        # x = self.dense(inputs)
        # x = self.dense2(x)
        x = layers.Dense(4)(inputs)
        x = layers.Dense(2)(x)

        return x


def main():

    model = MyModel()
    model.build(input_shape=(None, 50))
    model.summary()


if __name__ == "__main__":
    main()

问题出在call函数内部，不可以直接通过layers.Dense()的方式构建网络层，
应该在build函数内部先建立对应的层

正确代码应为：

import tensorflow as tf
from tensorflow.keras import layers, models


class MyModel(tf.keras.models.Model):
    def __init__(self):
        super(MyModel, self).__init__()

    def build(self, input_shape):
        self.dense = layers.Dense(4)
        self.dense2 = layers.Dense(2)
        # self.built = True
        super(MyModel, self).build(input_shape)

    def call(self, inputs):

        x = self.dense(inputs)
        x = self.dense2(x)
        # x = layers.Dense(4)(inputs)
        # x = layers.Dense(2)(x)

        return x


def main():

    model = MyModel()
    model.build(input_shape=(None, 50))
    model.summary()


if __name__ == "__main__":
    main()

同时如果只是通过self.built = True的方式，同样会出现此错误，建议通过super(MyModel, self).build(input_shape)的方式，使得build生效。

3. summary输出时output shape 为multiple的方法

在model.build()后手动调用call方法，并输入shape大小。
eg：

   model = MyModel()
   model.build(input_shape=(None, 50))
   model.call(tf.keras.Input(shape=(50, )))
   model.summary()

最终summary输出：

Model: "my_model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 4)                 204       
_________________________________________________________________
dense_1 (Dense)              (None, 2)                 10        
=================================================================
Total params: 214
Trainable params: 214
Non-trainable params: 0
_________________________________________________________________

Process finished with exit code 0