I've been working on a Team Fortress 2 market analyzer these holidays. The idea was that it would scrape tf2outpost (and later maybe tf2tp and other fora), determine the market price of items and then notify me of advantageous trades.
The scraping works fine, but the non-negative linear least squares solution gives wrong values (compared to tf2spreadsheet), even after trying to adjust for markup. I think the problem is that I have too little data (only ~2 times as many trades as items per day). The problem is parsing the free-form notes, since people often don't exactly offer the trade they put in.
This is the technology I used: emacs git CouchDB Python 2.7.2 Buildout html5parser lxml scipy.optimize.nnls