I haven't tried this myself but the procedure should be like this:
Install tensorflow by entering the following from a command line on Edison:
$ git clone https://github.com/tensorflow/tensorflow.git
$ cd tensorflow
$ make install
Tensorflow uses Bazel build which does not have binaries for Edison.
The first step is to install numpy - this works, numpy is compiled on target, on the fly.
if you run ./configure there is problem with sed: (using all default options or all No options)
Found possible Python library paths:
Please input the desired Python library path to use. Default is [/usr/lib/python2.7/site-packages]
Using python library path: /usr/lib/python2.7/site-packages
sed: unmatched '['
You can use tiny-dnn · GitHub instead [have not tested it yet]
This idea sounds absolutely awesome, and if there is any progress please let us know.
I was busy so I didn't check mail.
It seems tensorflow hasn't supported Intel Edison yet.
So i'm currently using tensorflow on my Raspberry pi 2.
tiny-dnn works on Edison 32bit
keras with theano backend also works on Edison 32bit