A Collection of Interesting Links

Deploying prototypes using docker Explains how to deploy prototypes to AWS uding docker. I used it for deploying a Flask app.
In How to build a data pipeline Balázs Kégl argues that one should focus on getting the data pipeline in production before doing any optimazation of the machine learning model. This is mainly because optimization is easy once everything else is in place.
In Fraud detection challenges and data skepticism using LIME Shir Meir Lador walks through a mistake she made when training a new machine learning model. The results looked promising as she beat the current model, but she began to feel suspicious about the results when she could do just as well using just two features. In her talk se describes how she figured out what the problem was. I made a similar mistake in my own work, so I found the talk both insightful and entertaining to watch.