Tensorflow is a code library of sophisticated machine learning algorithms. TensorFlow was created on, and intended for use on Linux machines. However, several weeks ago, a version of TensorFlow for Windows was released by the TF guys.
TensorFlow for Windows is pretty rough around the edges but I was able to get a fairly complex demo running. I went to tensorflow.org and clicked on the Install tab and found instructions for installing TF on Windows. The first requirement was Python 3.5 (you access the TF code library using Python).
I like Python, but the split between Python 2 and Python 3 is really, really irritating. I blew away my existing Python installation and then went to python.org and found a link to a self-extracting executable installer. The installation went smoothly.
Next, I edited my System environment PATH variable so that my shell commands would find python.exe and the pip3.exe installer.
Then I verified that Python was working by launching a Windows shell and issuing a python –version command. Then I installed TensorFlow for Windows.
(prompt) python --verion Python 3.5.3 (prompt) pip3 install --upgrade tensorflow . . . Successfully installed . . . (prompt)
Finally I verified TensorFlow by printing a Hello message:
>>> import tensorflow as tf >>> h = tf.constant('Hello message from TF!') >>> s = tf.Session() >>> print(s.run(h))
The shell spewed all kinds of warning messages but eventually printed my message. I will describe how I got a demo of image recognition using a convolutional neural network to run in a future blog post.
Bottom line: If you use Windows, TensorFlow for Windows is almost ready for mainstream use.