Summary
In this chapter, we looked at running jobs on big data. By most standards, our dataset is quite small—only a few hundred megabytes. Many industrial datasets are much bigger, so extra processing power is needed to perform the computation. In addition, the algorithms we used can be optimized for different tasks to further increase the scalability.
Our approach extracted word frequencies from blog posts, in order to predict the gender of the author of a document. We extracted the blogs and word frequencies using MapReduce-based projects in mrjob. With those extracted, we can then perform a Naive Bayes-esque computation to predict the gender of a new document.
We can use the mrjob library to test locally and then automatically set up and use Amazon's EMR cloud infrastructure. You can use other cloud infrastructure or even a custom built Amazon EMR cluster to run these MapReduce jobs, but there is a bit more tinkering needed to get them running.
