Why can I not import Tensorflow.contrib I get an error of No module named 'tensorflow.python.saved

Question

Hi I just installed Tensorflow on my Mac and I want to use tf.contrib.slim but when I use it I get this

import tensorflow as tf

slim = tf.contrib.slim

Error:

File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/tensorflow_estimator/python/estimator/export/export_lib.py", line 25, in from tensorflow.python.saved_model.model_utils import build_all_signature_defs ModuleNotFoundError: No module named 'tensorflow.python.saved_model.model_utils'

I don't know what to do, please help me

I use Tensorflow.13.1 and python 3.7

Answer 1

For anyone who is trying some old codes from github with Tensorflow 1.xx versions while having Tensorflow 2.0.x please note that tf.contrib no longer exist in Tensorflow 2.0.x and it's modules were moved.
Please google the name of the module without the tf.contrib part to know it's new location and thus migrating your code accordingly by correcting the import statement.

Hope this helped!

Answer 2

first unistall tensorflow

pip uninstall tensorflow

then install 1.13.2 version

pip install tensorflow==1.13.2

it works.. had the same issue.. but installing tensorflow 1.13.2 solved it!
the newer version of tensorflow doesn't have

Answer 3

first:

pip install --upgrade tf_slim

then:

import tf_slim as slim

Answer 4

I solved this by below method.

  pip uninstall tensorflow_estimator
  pip install tensorflow_estimator

the reference is : https://github.com/tensorflow/tensorflow/issues/27079

Answer 5

Currently the default install of tensorflow is 2.x while your code is for 1.x. The contrib module has been removed from tf 2.x. Check the warnings:

"The TensorFlow contrib module will not be included in TensorFlow 2.0"

Uninstall tensorflow and then install the 1.x version with

pip install tensorflow==1.15

Answer 6

If the following command doesn't work

pip3 install tensorflow==1.14.0 

then we can try the following command

pip3 install https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.14.0-py3-none-any.whl

Answer 7

tf.contrib has moved out of TF starting TF 2.0 alpha.

You can upgrade your TF 1.x code to TF 2.x using the tf_upgrade_v2 script https://www.tensorflow.org/alpha/guide/upgrade

Answer 8

I solved this using following steps:

First i do check current version of my tf using

import tensorflow
print(tensorflow.__version__)

2.5.0

Then contrib is part of older version of tensorflow as it removed from 2.X.

so we need to use tf version 1.X

that will be done using following snippet

%tensorflow_version 1.x
import tensorflow
print(tensorflow.__version__)

and you will get output

**TensorFlow 1.x selected.
1.15.2**

now you can use

from tensorflow.contrib import seq2seq
from tensorflow.contrib.rnn import DropoutWrapper

Answer 9

Contrib has been taken from tensorflow in 2.x version we have to switch to an older version.

If you are using in collab you can directly switch from 2.x version to 1.x with this command

%tensorflow_version 1.x

Doing this solved the problem.

Apart from collab you can just uninstall your current version and install 1.15.2 with the following commands.

pip uninstall tensorflow

pip install tensorflow==1.15.2

Answer 10

As mentioned in the answers, tensorflow.contrib is not supported in tensorflow 2. I fixed the problem with creating an environment in conda with its python version set to 3.7 and setting tensorflow version to 1.14. You may face a one or two bugs related to package compatibility but eventually it'll work.

Answer 11

In GoogleColab try:

%tensorflow_version 1.x

and

!pip install q keras==2.0.5

then restart run time.

Answer 12

I think majority of solution is suggesting to downgrade tensorflow version. At first place why does TF 2 does not support WALS factorization?One blog suggest its because official recommendataion is NCF Matrix Factorization in tensorflow 2.0 using WALS Method .

Answer 13

Editing the code may be a bit tedious but if you are willing to make the effort - just add this in all tensorflow links

compat.v1

so tf.assign -> tf.compat.v1.assign

etc

Answer 14

As suggested in the Migration Guide from TensorFlow 1.x to TensorFlow 2 page , install TF-Slim , available on Google-research Github page and then use

import tf_slim as slim

instead of tf.contrib.layers

Answer 15

Starting from tensorflow 13.1 there's no contrib. You can use without it slim = tf.slim or you can install pip install tensorflow==1.13 and use it

Answer 16

for running it on python3 I used pip3 to install

pip3 install tensorflow

This worked for me