Keras import loading neural network - python-3.x

Question

The script normally works on a local machine, but it also fails when it is uploaded to the server. The program falls on the line:
from keras.models import model_from_json
All versions of libraries on a laptop and server are the same. Python 3.6.5, Keras 2.2.4, Tensorflow 1.5.0, Numpy 1.14.3
Error message:
(base) C:\classX5>python app.py
Using TensorFlow backend.
Traceback (most recent call last):
File "app.py", line 8, in <module>
from keras.models import model_from_json
File "C:\Anaconda3\lib\site-packages\keras\__init__.py", line 3, in <module>
from . import utils
File "C:\Anaconda3\lib\site-packages\keras\utils\__init__.py", line 27, in <mo
dule>
from .multi_gpu_utils import multi_gpu_model
File "C:\Anaconda3\lib\site-packages\keras\utils\multi_gpu_utils.py", line 7,
in <module>
from ..layers.merge import concatenate
File "C:\Anaconda3\lib\site-packages\keras\layers\__init__.py", line 4, in <mo
dule>
from ..engine.base_layer import Layer
File "C:\Anaconda3\lib\site-packages\keras\engine\__init__.py", line 8, in <mo
dule>
from .training import Model
File "C:\Anaconda3\lib\site-packages\keras\engine\training.py", line 21, in <m
odule>
from . import training_arrays
File "C:\Anaconda3\lib\site-packages\keras\engine\training_arrays.py", line 8,
in <module>
from scipy.sparse import issparse
File "C:\Anaconda3\lib\site-packages\scipy\__init__.py", line 119, in <module>
from scipy._lib._ccallback import LowLevelCallable
File "C:\Anaconda3\lib\site-packages\scipy\_lib\_ccallback.py", line 1, in <mo
dule>
from . import _ccallback_c
ImportError: cannot import name '_ccallback_c'

Answer 1

All versions of libraries on a laptop and server are the same. Python
3.6.5, Keras 2.2.4, Tensorflow 1.5.0, Numpy 1.14.3
But NOT the scipy package.
Just uninstall (pip3 uninstall scipy) and reinstall scipy package. It solved the problem for many people.

Answer 2

Posting my same answer here I posted on the linked question:
This is likely an issue with the SciPy Python module compatibility you have installed in your site-packages folder and the OS architecture you're running it on.
After digging in, to give the full background on this, first of all SciPy relies on having NumPy already installed. The SciPy wheel's setup.py file uses NumPy functionality to configure and install the wheel.
SciPy setup.py:
...
if __name__ == '__main__':
from numpy.distutils.core import setup
setup(**configuration(top_path='').todict())
Secondly, when just trying to use the wheel, if you run into this error, you can see after inspecting the wheel's files that the reason is the binary wheels have a naming convention where the shared object file, here it's called _ccallback_c.so, is instead named based on the architecture that the binary wheel supports. When trying to import the shared object by file name in /_lib/_ccallback.py it can't find it, hence this error (line 1 in /_lib/_ccallback.py) because, instead of being named _ccallback_c.so it's called _ccallback_c.cpython-36m-x86_64-linux-gnu.so or another architecture variation:
from . import _ccallback_c
These file names may be an artifact of having not run the NumPy setup process yet or something related to Cython, that I'm not quite sure about. But the easiest fix is to change the .whl extension to .zip and rename all those relevant .so files to not contain the architecture snippet. Then change .zip -> .whl and it should be good to go.