fix encoding issue in word2vec example

[amygdala]

I added a note earlier but figured I should make it a proper issue. Can you revisit this commit: https://github.com/amygdala/tensorflow-workshop/commit/5393ac52e6bded922d66789412c9b451bcae42da#diff-300e5b2cbe160af1a9b56b074a6e6b04R144 which now works for py 3 but not py 2.7? (whereas reverting the commit lets it work for py 2.7 but not py 3...)