The Offical Installation Guide has not mentioned any instruction about how to install MXNet on Fedora/RedHat/CentOS.

This is because some dependencies library names are different compared to Ubuntu/Debian.

I installed MXNet successfully on Fedora 23 and wrote the guide to help me remember the installation steps.

1.Install dependencies

dnf install opencv-devel atlas-devel 
# make sure /usr/bin/ld  can find these library
cp /usr/lib64/atlas/* /lib64/ 

You may also need to install gcc, git and other build tools

2.change the compilation command

If you follow the Official Guide, you may get an error when doing ‘make’ files.

/usr/bin/ld: cannot find -lcblas

This is because names and content of atlas libraries have changed.

chemal:

Names and content of atlas libraries have changed recently. Try

-L/usr/lib64/atlas -lsatlas or -L/usr/lib64/atlas -ltatlas

instead of -lcblas.

So, We need change the complication command from -lcblas to -lsatlas or -ltatlas at mshadow/make/mshadow.mk, line 57

For me, I use -lsatlas:

MSHADOW_LDFLAGS += -lsatlas 

3.make -j4

Then you can make files successfully.

4.Try Python version with CPU

I try using example/neural-style/run.py to draw a picture.

First, run download.sh to download the model and input files.

The script need import kitimage package.

pip install scikit-image

Install Python package as the guide mentioned.

I always install package system-widely, which requires root permission.

cd python; sudo python setup.py install
export PYTHONPATH=~/mxnet/python

Finally, run command

python run.py --content-image input/1.jpg --style-image input/starry_night.jpg --gpu -1 ##using CPU

You can get a neural style image. I post the SDTOUT INFO at gist.