Feedback request. I implemented the first example on Probabilistic programming for hackers: Inferring behaviour from text message data

1

I attempted to translate the example Inferring behaviour from text message data.
This is the link to the code.

Honestly, I don’t find my code satisfactory. I’m looking to read your feedback.
In particular, I didn’t know how to model this:
27%20PM
where t is different for each observation.
Was a tf.placeholder is required?

EDIT
It works now.