I'd like to explore 'data science'. The term seems a little vague to me, but I expect it to require:
What are some good datasets and problems, accessible to a statistician with some programming background, that I can use to explore the field of data science?
To keep this as narrow as possible, I'd ideally like links to open, well used datasets and example problems.
Just head to kaggle.com; it'll keep you busy for a long time. For open data there's the UC Irvine Machine Learning Repository. In fact, there's a whole Stackexchange site devoted to this; look there.
The Sunlight Foundation is an organization that is focused on opening up and encouraging non-partisan analysis of government data.
There is a ton of analysis out there in the wild that can be used for comparison, and a wide variety of topics.
They provide tools and apis for accessing data, and have helped push to make data available in places like data.gov.
One interesting project is Influence Explorer. You can get source data here as well as access to real time data.
You might also want to take a look at one of our more popular questions:
Is your Masters in Computer Science? Statistics?
Is 'data science' going to be at the center of your thesis? Or a side topic?
I'll assume your in Statistics and that you want to focus your thesis on a 'data science' problem. If so, then I'm going to go against the grain and suggest that you should not start with a data set or an ML method. Instead, you should seek an interesting research problem that's poorly understood or where ML methods have not yet been proven successful, or where there are many competing ML methods but none seem better than others.
Consider this data source: Stanford Large Network Dataset Collection. While you could pick one of these data sets, make up a problem statement, and then run some list of ML methods, that approach really doesn't tell you very much about what data science is all about, and in my opinion doesn't lead to a very good Masters thesis.
Instead, you might do this: look for all the research papers that use ML on some specific category -- e.g. Collaboration networks (a.k.a. co-authorship). As you read each paper, try to find out what they were able to accomplish with each ML method and what they weren't able to address. Especially look for their suggestions for "future research".
Maybe they all use the same method, but never tried competing ML methods. Or maybe they don't adequately validate their results, or maybe there data sets are small, or maybe their research questions and hypothesis were simplistic or limited.
Most important: try to find out where this line of research is going. Why are they even bothering to do this? What is significant about it? Where and why are they encountering difficulties?
