Keras model.to_json() error: 'rawunicodeescape' codec can't decode bytes in position 94-98: truncated \uXXXX

Question

model.to_json()

for the model

____________________________________________________________________________________________________ Layer (type) Output Shape Param #
Connected to
==================================================================================================== lambda_1 (Lambda) (None, 3, 160, 320) 0
lambda_input_1[0][0]
____________________________________________________________________________________________________ convolution2d_1 (Convolution2D) (None, 1, 40, 16) 327696
lambda_1[0][0]
____________________________________________________________________________________________________ elu_1 (ELU) (None, 1, 40, 16) 0
convolution2d_1[0][0]
____________________________________________________________________________________________________ convolution2d_2 (Convolution2D) (None, 1, 20, 32) 12832
elu_1[0][0]
____________________________________________________________________________________________________ elu_2 (ELU) (None, 1, 20, 32) 0
convolution2d_2[0][0]
____________________________________________________________________________________________________ convolution2d_3 (Convolution2D) (None, 1, 10, 64) 51264
elu_2[0][0]
____________________________________________________________________________________________________ flatten_1 (Flatten) (None, 640) 0
convolution2d_3[0][0]
____________________________________________________________________________________________________ dropout_1 (Dropout) (None, 640) 0
flatten_1[0][0]
____________________________________________________________________________________________________ elu_3 (ELU) (None, 640) 0
dropout_1[0][0]
____________________________________________________________________________________________________ dense_1 (Dense) (None, 512) 328192
elu_3[0][0]
____________________________________________________________________________________________________ dropout_2 (Dropout) (None, 512) 0
dense_1[0][0]
____________________________________________________________________________________________________ elu_4 (ELU) (None, 512) 0
dropout_2[0][0]
____________________________________________________________________________________________________ dense_2 (Dense) (None, 1) 513
elu_4[0][0]
==================================================================================================== Total params: 720,497 Trainable params: 720,497 Non-trainable params: 0 ____________________________________________________________________________________________________ None

throws the exception

'rawunicodeescape' codec can't decode bytes in position 94-98: truncated \uXXXX

What could be the problem and how can I solve it?

Answer 1

I came across a similar problem when using keras 1.2.1 with a tensorflow-gpu backend.

I found out it was caused because windows 10 anniversary edition was having problems encoding the forward slash character.

Using the Lambda layer makes the to_json() call fail but switching to batch normalization works just fine.

model = Sequential()

# model.add(Lambda(lambda x: x / 255. - .5, input_shape=INPUT_DIMENSIONS))
model.add(BatchNormalization(input_shape=INPUT_DIMENSIONS, axis=1))
. . . 
# POST PROCESSING, SAVE MODEL TO DISK
with open('model.json', 'w') as json_file:
    json_file.write(model.to_json())

Not an ideal solution but hopefully it works for someone looking at this in the future.

Answer 2

It seems like your code is in directory like this: "C:\Users\python\u{...}.py". This kind of error is related to python 3 version, where we get special character \u and can't decode it on Windows machines. You can change filename or full path to file, so that it did not contain special characters or make a patch to function 'func_dump' from file generic_utils.py (it can be reached by following path 'keras/utils/generic_utils.py'). You should replace line code = marshal.dumps(func.__code__).decode('raw_unicode_escape') by line code = marshal.dumps(func.__code__).replace(b'\\',b'/').decode('raw_unicode_escape').

Answer 3

I believe that this is actually due to a bug in Keras, and was thankfully fixed by Keras pull request #8572. However, that PR was only merged very recently, and while it's incorporated in the latest pypi release, it's not in the latest conda release, which is why I was getting this error.

Installing Keras via pip install keras instead of conda install keras made everything work for me (though I also had to precede this with conda install pip to get pip to work properly).