Python Keras module ' keras.backend ' has no attribute ' Image_data_format '

Problem:

When you run the sample program MNIST_CNN with Keras, the following error occurs: ' Keras.backend ' has no attribute ' Image_data_format '

Program Path https://github.com/fchollet/keras/blob/master/examples/mnist_cnn.py

The Python Conda environment used is the carnd-term1 of the Udacity autopilot course

Fault Program segment:

if ' Channels_first ' :     = X_train.reshape (x_train.shape[0], 1, Img_rows, img_cols)    = X_test.reshape (x_test.shape[0], 1, Img_ Rows, img_cols)    = (1, img_rows, img_cols)else:    = X_train.reshape (x_ Train.shape[0], Img_rows, Img_cols, 1)    = X_test.reshape (x_test.shape[0], img_rows, Img_cols, 1) c17/>= (img_rows, img_cols, 1)

Complete code See https://github.com/fchollet/keras/blob/master/examples/mnist_cnn.py

Fault Analysis:

The version of the Keras version in the Conda environment is old in the subroutine, so there is no ' image_data_format ' variable

Workaround:

Here are two ways to choose one

1) If the Keras version is not upgraded

将 K.image_data_format() == ‘channels_first‘Replaced byK.image_dim_ordering() == ‘th‘

2) Upgrade the Keras version to the latest

> Activate Carnd-term1//activating your Conda environment, this environment of mine is called Carnd-term1

(CARND-TERM1) > Conda list//show packages installed in the current environment

(CARND-TERM1) > Pip uninstall Keras//Uninstall old version of Keras, Keras is installed using PIP, so both uninstall and reload use PIP related commands

(CARND-TERM1) > Pip install Keras//Reinstall new version of Keras

(CARND-TERM1) > Conda list//check if the version has been updated

Python Keras module ' keras.backend ' has no attribute ' Image_data_format '