It’s been a few years since I’ve last commented on my deployed Kubernetes-based workbench. Over that time I’ve learned a lot and gotten more experience with it and other solutions - it’s time to revisit the setup and decide whether to upgrade or replace.

View comments.

A reader recently pointed out that I hadn’t actually tested math display in Jupyter (Thanks Ben!). Here are the steps I had to take to get that working…

View comments.

Before we can begin in earnest there are just a few requirements we need to put together, but many of you will already have these in place - an AWS account, a github account, and a domain name. If you don’t we’ll have you sorted out in just a …

View comments.

Now we’ll start getting to the fun part - actually deploying our infrastructure and launching a basic jupyterhub deployment. While I’ve simplified many steps, I’ve left many other intentionally exposed - to offer opportunities for customization and to expose how things work. What follows is a condensed summary of …

View comments.

We’ve sketched out a rough architecture at this point, time to get an estimate of expected cost, and think through how well we can expect it to meet our original design goals. Besides overall user experience, I find it’s important to carefully think through the limitations of any …

View comments.

Now that we have some of our design goals in place for the new environment we can start evaluating our options and creating the solution architecture. It will be critical to balance cost, performance, ease of administration, and ease of use in order to make this a truly viable platform …

View comments.

One of the most daunting tasks of getting going on a personal data science project can be just getting started with the environment. What I generally want is instant access to the latest technology stack (hopefully at low cost), allowing me to get to work quickly while also exploring the …

View comments.

This year I’ve had the pleasure of participating in a team at MIT identifying characteristics that make patients likely to be intensive care unit “Frequent Fliers” - patients with multiple ICU admissions in a short time span. This series will explore the implementation of text-based patient phenotyping we use as …

View comments.

I’ve recently been upgrading my tool set to the latest versions of Python, Keras, and Tensorflow, all running on a docker-based GPU-enabled deployment of Jupyter with CUDA 8 installed. In this notebook I walk through a quick test to verify that the GPU is recognized and is working …

View comments.

We saw in the previous post that there is promise to using ICD9 codes for pre-classifying encounters more likely to have our concepts of interest. In this post we’ll walk through building simple logistic regression classifiers based on a training data set, and will evaluate their performance on a …

View comments.