Best

Cutting edge data science projects.

Deep Learning frameworks such as Theano, Caffe, TensorFlow, Torch, MXNet and CNTK are the work horses of Deep Learning work. These frameworks as well as the GPU (predominantly Nvidia) are the what enables the rapid growth of Deep Learning. It was refreshing to hear Nando de Freitas acknowledge their work in the recently concluded NIPS 2016 conference. Infrastructure does not get enough of the recognition it deserves in the academic community. Yet, programmers toil on to continually tweak and improve their frameworks.

Recently, a new framework was revealed by Facebook and a bunch of other partners (Twitter * NVIDIA * SalesForce * ParisTech * CMU * Digital Reasoning * INRIA * ENS). PyTorch came out of stealth development. PyTorch is an improvement over the popular Torch framework (Torch was a favorite at DeepMind until TensorFlow came along). The obvious change is the support of Python over the less often used Lua language. Almost all of the more popular frameworks use Python, so it is a relief that Torch has finally joined the club.

Betaworks recently announced that they had collected data from over 40,000 users who shared their iPhone homescreens through their apptly named #homescreen app. As they stated in their announcement, the apps people keep on their homescreen are often the apps they use the most. Being a data-minded individual I thought, “I wonder what kind of questions you could ask of this kind of data?” Of course, I immediately jumped to using the data to try and understand the landscape of self-tracking and quantified self app use. Let’s dive in.

What does “causality” mean, and how can you represent it mathematically? How can you encode causal assumptions, and what bearing do they have on data analysis? These types of questions are at the core of the practice of data science, but deep knowledge about them is surprisingly uncommon.

If you analyze data without regard to causality, you open your results up for the possibility of enormous biases. This includes everything from recommendation system results, to post-hoc reports on observational data, to experiments run without proper holdout groups.

Recent posts have been aimed at a more general audience. This one will be aimed at practitioners, and will assume a basic working knowledge of math and data analysis. To get the most from this post you should have a reasonable understanding of linear regression and probability (although we’ll review a lot of probability). Prior knowledge of graphical models will make some concepts more familiar, but is not required.