AttributeError: module ‘tensorflow.keras. utils‘ has no attribute image_ dataset_ from_ Directory - solution

In raspberry pie , by python3.7 Set up well tensorflow2.4（ Brief process of construction ） After the environment of , There are problems like this when running code
First , utilize ：

import tensorflow as tf
print(tf.keras.utils.__file__)

Search for tf.keras.utils Modular __init__.py Path to file ：

then , Open this tensorflow/keras/utils/__init__.py file , Find out tensorflow.keras.utils The functions and classes in are from tensorflow.python.keras in import Over here , After with windows Next tensorflow 2.7 Of __init__.py Contrast .

windows Of __init__.py The documents are as follows ：

On raspberry pie, follow ：

from tensorflow.python.keras.preprocessing.image_dataset import image_dataset_from_directory

I found my path image_dataset_from_directory：

It means raspberry pie tf There is this function , It's just tensorflow/keras/utils/__init__.py No document import This function , So at runtime, there is no __init__.py This function is found in the file and an error is reported .

You can see ,tensorflow/keras/utils/__init__.py Files are automatically generated , Not necessarily complete ：

So add a pair of image_dataset_from_directory Function reference

Then add ：

from tensorflow.python.keras.preprocessing.image_dataset import image_dataset_from_directory

perhaps , Directly in tensorflow/keras/utils/__init__.py Add... To the file ：

from tensorflow.python.keras.preprocessing.image_dataset import image_dataset_from_directory

after , It can be used normally in the code tf.keras.utils.image_dataset_from_directory 了

notes ： Similar problems may also be solved through this method of viewing the source code .
such as , The same version of tensorflow Of tf.lite The next level is funcA This function , But in the new environment AttributeError, You can start with ：

import tensorflow as tf
print(tf.lite.__file__)

to glance at tf.lite Of __init__.py Path to file ：

Then similar to the above method , And can quote normally funcA Functional tensorflow Of __init__.py Contrast , In the new environment __init__.py Add... To the file funcA Function related references are ok