Hadoop Best Practices

In the course of my day-to-day work, I’ve had the opportunity to work with home-grown Map-Reduce frameworks as well as Hadoop. On the whole, I can say that I’m really pleased with the stability that Hadoop provides. As such, I’ve had the opportunity to talk at a few venues about Hadoop and map reduce related things.

We at Explorys are trying very hard to build up a vibrant big-data community in Cleveland. Part of that effort surrounds the care and feeding of a user group. Doing my part, I spoke recently about the set of best practices that I recommend based on my experiences with Hadoop.

In the course of preparing the slide deck, I wanted to demonstrate a fully working, self-contained example demonstrating my preferred methods to testing and designing a map reduce job. Also, I wanted the job to be slightly more complex than the canonical word-count example, yet not so complex that the average developer would not be able to wrap their head around it in the course of an hour.

I ended up going with doing an analysis of a smallish dataset that I keep of senatorial speeches from the 111th congress. I wanted to, particularly, investigate the words that were most “important” for each political grouping. In order to do that, I computed the IDF ranking function for each word and trimmed the top-k words according to that ranking function.

Casey Stella 20 March 2012 Cleveland, OH