[AttributeError] module ‘tensorflow’ has no attribute ‘Session’


2020-10-25 Python TroubleShooting

When I was trying the exercise of TensorFlow Programming Concepts, an error “AttributeError module ‘tensorflow’ has no attribute ‘Session'” occurred.

Error Detail

The code run in Google Colaboratory is the following.Generic HighlightingEnlighterJS Syntax Highlighter

This image has an empty alt attribute; its file name is 38dbab30e4232a9115cd0f972a88d428.png
An attribute error

Cause

tf.Session can be used in tensorflow 1.x version, while it cannot be used in 2.x version.

Condinf style of TensorFlow ver 2.x are different from ver 1.x. Version 2.x is more pythonic than 1.x. Please check “Effective TensorFlow 2” for details.

Checklist

  • Check if the version of tensorflow imported is 1.x or not.

Solution

solution 1. Change the version of tensorflow to 1.x

In Google Colaboratory, you can change the version of tensorflow with “%tensorflow_version” as below.Generic HighlightingEnlighterJS Syntax Highlighter

This method doesn’t work after importing tensorflow once, so restart runtime and change the version before import tensorflow.

solution 2. Use tf.compat.v1.Session()

Use tf.compat.v1.Session() instead of tf.Session().
Reference: Migrate your TensorFlow 1 code to TensorFlow 2Generic HighlightingEnlighterJS Syntax Highlighter