Sunday, June 11, 2017

Instaling TensorFlow on mac OSX Yosemite

I wanted to try out TensorFlow, so I started with the installation steps. First create virtualenv, and then try to do pip install, there are bunch of different ways you can install as per the instructions provided here, though I went with below and it at least for the first part installed it right.

pip install --ignore-installed --upgrade \

Above is CPU-only installation with Anaconda. My mac default python runs under anaconda, so above worked well for me compared to other installation mechanism described in documentation.

After the installation I followed below step to try out test of tensorflow installation success -

$ python
>> import tensorflow as tf

Though I ran into error -
    _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description) 
ImportError: numpy.core.multiarray failed to import

It was because of the numpy, it was pointing to the wrong one. I tried below to check the path of numpy -

$ python
>> import numpy
>> print numpy.__path__

Which is an issue, as its using /local/lib/python2.7 installation and not local virtualenv installation -
Go to the path and remove the file explicitly. After that I tried again to review the path of numpy -

>> print numpy.__path__

However, at this point I run into another error -

    from google.protobuf import descriptor as _descriptor
ImportError: No module named protobuf

Its because of issue with protobuf installation version, after looking at few of the online documentation help,  I updated to below -

$ pip uninstall protobuf
$ pip install protobuf==3.0.0a3

though error continues, 

    from google.protobuf import any_pb2 as google_dot_protobuf_dot_any__pb2
ImportError: cannot import name any_pb2

so figured another version from online help -

$ pip install --upgrade protobuf==3.0.0b2

After that it started working fine. -

>>> import tensorflow as tf
>>> hello = tf.constant("Hello, TensorFlow")
>>> sess = tf.Session()
2017-06-11 09:45:35.310965: W t
>>> print(
Hello, TensorFlow

>>> node1 = tf.constant(3.0, tf.float32)
>>> node2 = tf.constant(4.0)
>>> print(node1, node2)
(, )

If I do more experimentation around it, will share more examples and issues I run into and how to solve..