Ghetto by Mitchell Duneier

Tuesday, April 12, 2016

About 5 years ago I was a researcher at Princeton and worked with Mitchell Duneier. We used historical text analysis to evaluate his thesis, it’s great to see his book in print. To do part of the analysis I used Google’s text n-gram data and a mix of Python, awk, bash, and R scripts.


Here’s a New York Time review with more details on his work.

Map Your World and Ona at Geo for Good 2015

Wednesday, November 25, 2015

This past October I spoke on behalf of Map Your World at the 2015 Geo for Good User Summit.

Map Your World empowers youth to explore issues and ideas that matter - like clean drinking water, or food justice – then write surveys, collect data, and create maps to make change in their communities.

Map Your World is powered by the Ona API.

Below is a video of the talk I gave, which includes a clip from the film The Revolutionary Optimists.

Writing Python Code to Decide an Election

Friday, October 03, 2014

Yesterday I spoke at PyConZA 2014 about Ona’s work building the vote tallying system for the Libyan Constitutional Assembly Election last February.

The slides from my talk are below:

Here is the abstract:

Earlier this year Ona was given three weeks to write the software that will tally votes in the Libyan elections and decide who wins and who loses. This is not something we could get wrong. We combined agile development with best practices in testing and QA to build an open source tally system that was well tested, accurate, and easy to use. We will describe a success story of iterative behavior/test-driven-development under extreme conditions. Did the structure of the data change the day before the election? Yes. Did we have the tests to ensure that our implementation changes would not compromise the system’s integrity? Yes, and they didn’t.

This talk provides a narrative to both Software Engineers and Tech/Product Managers describing why best practices are essential for any organization and any project of any size. We will provide the audience with:

Real world examples they can implement in their own workflow and organizations, Insight into what succeeded (quick iteration with prioritization) and what was challenging (nothing being static), Anecdotes and coherent arguments they can take back to their organization to advocate for best practices.

Below is the full video of my presentation:

Automated Infrastructure with Pallet and Clojure

Monday, May 05, 2014

At Ona we are rebuilding our data management platform. We are starting with a light weight front-end that will serve up content pulled from the REST API of our current application. We are aiming to have the back-end in Clojure, the front-end in ClojureScript, and the infrastructure in Clojure using Pallet. We are excited to have a single (and a great) language handle all of these responsibilities.

We are still at a very early stage but we are a distributed team and like to have our apps on development boxes as we go. This allows us to share a common reference point, give mini-demos, and QA each other’s changes. Like Fabric for Python and Capistrano for Ruby, Pallet let’s us do quick deploys of the latest master or branch code.

Even better, Pallet let’s us write Clojure to bring up new clusters, similarly to Puppet, Chef, or Ansible – but in Clojure. We deploy to EC2 on AWS and are glad to avoid spending time mucking around in the AWS GUI. A succinct pallet file specifies the instance, the web application, and the deployment. Putting the current code online and bringing up a server (if one doesn’t already exist) is a single command:

lein do uberjar, with-profile +pallet pallet up \
--phases install,configure,deploy

This tells Leiningen to first create an uberjar, which puts all of our app’s dependencies in a single jar file. It then uses the pallet profile to install, configure, and deploy our application. This command is idempotent, making it easy to push the latest jar up.

A nuance we did not anticipate is that you cannot output logs to stdout in a Jetty app. This is not particularly surprising, but using stdout was a development configuration that we had not yet bothered to abstract.

For now we are handling this with the below middleware wrapper:

(defn wrap-with-logger [handler verbose?]
  (if verbose?
      (logger/wrap-with-logger handler "/dev/stdout")
          (fn [request] (handler request))))

This does the normal logging if verbose? is true and otherwise does nothing. When you run lein ring server-headless a handler is called which sets verbose? to true. When you run the app through java -jar ..., as in our pallet configuration, verbose? is set to false.

The ona-viewer project is a work-in-progress and we would welcome any feedback. Check it out on github.

Cascalog at Intent Media

Monday, April 28, 2014

While I was at Intent Media I led the data engineering team in rebuilding and extending the Intent Media data platform. To structure and simplify queries we relied on Cascalog, a Clojure DSL built on top of the Cascading library that is built on top of Apache Hadoop.

Cascalog is inspired by Datalog and uses logic programming to simplify query expression. It is similar to Datomic for Clojure and the recent DataScript for ClojureScript. This allows simple and concise queries, e.g. to compute the average age per country:

(?<- (stdout) [?country ?avg] 
   (location ?person ?country _ _) (age ?person ?age)
   (c/count ?count) (c/sum ?age :> ?sum)
   (div ?sum ?count :> ?avg))

Jon Sondag, a data scientist at Intent Media, recently gave a presentation at the NYC Clojure Meetup about Cascalog in production. His slides are embedded below.

It is great to see Cascalog being used in production data platforms.