You can use the query game API endpoint (passing in a parameter that lets you retrieve the entire game history). You'll have enough information to represent maps as graphs (with territories are vertices) and track player movements through them.
If you're working in Python, I wrote a quick wrapper library for the API a while ago: https://github.com/knyte/wl_api
It does take some time to scrape large numbers of games, though, but then again so do most worthwhile things in data science.
Also, it might be cool to scrape 1v1 ladder games (using the getGameIDs function; keep in mind that the template used on the ladder has changed over time). I bet you might get some interesting information on player behavior out of them, maybe by running some clustering algorithms to detect pick combinations that go together.
EDIT: Another idea, but this might take a while: query the x most recent WL games (just iterating through game IDs and only accepting real, finished games) and using a clustering algorithm, identify Warlight's "social networks" based on who plays with whom. This would make for some pretty amazing dataviz tbh.
Edited 2/20/2017 23:46:46