How to fix "AttributeError: module 'tensorflow' has no attribute 'get_default_graph'"?

I am trying to run some code to create an LSTM model but i get an error:

AttributeError: module 'tensorflow' has no attribute 'get_default_graph'

My code is as follows:

from keras.models import Sequential model = Sequential() model.add(Dense(32, input_dim=784)) model.add(Activation('relu')) model.add(LSTM(17)) model.add(Dense(1, activation='sigmoid')) model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) 

I have found someone else with a similar problem and they updated tensorflow and it works; but mine is up to date and still does not work. I am new to using keras and machine learning so I apologise if this is something silly!

19 Answers

Please try:

from tensorflow.keras.models import Sequential

instead of

from keras.models import Sequential

For tf 2.1.0 I used tf.compat.v1.get_default_graph() - e.g:

import tensorflow as tf sess = tf.compat.v1.Session(graph=tf.compat.v1.get_default_graph(), config=session_conf) tf.compat.v1.keras.backend.set_session(sess) 

for latest tensorflow 2 replace the above code with below code with some changes

for details check keras documentation:

import tensorflow as tf from tensorflow import keras from tensorflow.keras import layers from tensorflow.keras.models import Sequential, load_model model = tf.keras.Sequential() model.add(layers.Dense(32, input_dim=784)) model.add(layers.Activation('relu')) model.add(layers.LSTM(17)) model.add(layers.Dense(1, activation='sigmoid')) model.compile(loss='binary_crossentropy', optimizer=tf.keras.optimizers.Adam(0.01), metrics=['accuracy']) 

It occurs due to changes in tensorflow version :: Replace

tf.get_default_graph() 

by

tf.compat.v1.get_default_graph() 

I had the same problem. I tried

from tensorflow.keras.models import Sequential 

and

from keras.models import Sequential 

none of them works. So I update tensorflow, keras and python:

$conda update python $conda update keras $conda update tensorflow 

or

pip install --upgrade tensorflow pip install --upgrade keras pip install --upgrade python 

My tensorflow version is 2.1.0; my keras version is 2.3.1; my python version is 3.6.10. Nothing works until I unintall keras and reinstall keras:

pip uninstall keras pip install keras --upgrade 

Turns out I was using the wrong version (2.0.0a0), so i reset to the latest stable version (1.13.1) and it works.

Replace all keras.something.something with tensorflow.keras.something, and use:

import tensorflow as tf from tensorflow.keras import backend as k 

Downgrading will fix the problem but if you want to use latest version, you must try this code: from tensorflow import keras and 'from tensorflow.python.keras import backend as k That's work for me

Use the following:

tf.compat.v1.disable_eager_execution() print(tf.compat.v1.get_default_graph()) 

It works for tensorflow 2.0

YES, it won't work since you are using the updated version of tensorflow ie tensorflow == 2.0 , the older version of tensorflow might help. I had the same problem but i fixed it using the following code.

try:

import tensorflow as tf from tensorflow import keras from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense from tensorflow.keras.layers import Dropout 

instead:

from keras.models import Sequential from keras.layers import Dense from keras.layers import Dropout 

To solve the problem I used the code below:

from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense import numpy 

!pip uninstall tensorflow !pip install tensorflow==1.14 

this worked for me... working on hrnetv2.. ty

This worked for me. Please use the below import

from tensorflow.keras.layers import Input 

This has also happend to me. The reason is your tensorflow version. Try to get older version of tensorflow. Another problem can be you have a python script named tensorflow.py in your project.

Yes, the code is not working with this version of tensorflow tensorflow == 2.0.0 . moving to version older than 2.0.0 will help.

Assuming people referring to this thread will be using more and more tensorflow 2:

Tensorflow 2 integrates further keras api, since keras is designed/developed very wisely. The answer is very easy if you are using tensorflow 2, as described also here:

from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, Activation, LSTM model = Sequential() model.add(Dense(32, input_dim=784)) model.add(Activation('relu')) model.add(LSTM(17)) model.add(Dense(1, activation='sigmoid')) model.compile(loss=tensorflow.keras.losses.binary_crossentropy, optimizer=tensorflow.keras.optimizers.Adam(), metrics=['accuracy']) 

and that's how you change one would use something like MNIST from keras official page with just replacing tensorflow.keras instead of keras and runnig it also on gpu;

from __future__ import print_function import tensorflow from tensorflow.keras.datasets import mnist from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, Dropout, Flatten from tensorflow.keras.layers import Conv2D, MaxPooling2D from tensorflow.keras import backend as K batch_size = 1024 num_classes = 10 epochs = 12 # input image dimensions img_rows, img_cols = 28, 28 # the data, split between train and test sets (x_train, y_train), (x_test, y_test) = mnist.load_data() if K.image_data_format() == 'channels_first': x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols) x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols) input_shape = (1, img_rows, img_cols) else: x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1) x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1) input_shape = (img_rows, img_cols, 1) x_train = x_train.astype('float32') x_test = x_test.astype('float32') x_train /= 255 x_test /= 255 print('x_train shape:', x_train.shape) print(x_train.shape[0], 'train samples') print(x_test.shape[0], 'test samples') # convert class vectors to binary class matrices y_train = tensorflow.keras.utils.to_categorical(y_train, num_classes) y_test = tensorflow.keras.utils.to_categorical(y_test, num_classes) model = Sequential() model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=input_shape)) model.add(Conv2D(64, (3, 3), activation='relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.25)) model.add(Flatten()) model.add(Dense(128, activation='relu')) model.add(Dropout(0.5)) model.add(Dense(num_classes, activation='softmax')) model.compile(loss=tensorflow.keras.losses.categorical_crossentropy, optimizer=tensorflow.keras.optimizers.Adadelta(), metrics=['accuracy']) model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=1, validation_data=(x_test, y_test)) score = model.evaluate(x_test, y_test, verbose=0) print('Test loss:', score[0]) print('Test accuracy:', score[1]) 

For TensorFlow 2.0, use keras bundled with tensorflow.

try replacing keras.models with tensorflow.python.keras.models or tensorflow.keras.models:

from tensorflow.python.keras.models import Sequential from tensorflow.python.keras.layers.core import Dense, Activation 

This should solve the problem.

To resolve version issues in TensorFlow, it's a good idea to use this below technique to import v1 (version 1 or TensorFlow 1. x) and we also can disable the TensorFlow 2. x behaviors.

import tensorflow.compat.v1 as tf tf.disable_v2_behavior() 

You can refer to the following link to check the mapping between Tensorflow 1. x and 2. x

Please try to be concise!

First -->

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

Then -->

model = keras.Sequential( [ layers.Dense(layers.Dense(32, input_dim=784)), layers.Dense(activation="relu"), layers.Dense(LSTM(17)) ] ) model.add(layers.Dense(1, activation='sigmoid')) model.compile(loss='binary_crossentropy', optimizer=tf.keras.optimizers.Adam(0.01), metrics=['accuracy']) 

and voila!!