Has anyone here been using the Google Prediction API? For what? Did it "work"?
If you're looking for a real-world case check out this one which automatically assigns priority to bugs.
I've also created a couple test projects to get a feel for how I might use it in production.
One of them suggests tags for Stackoverflow questions. The model for this was trained with questions (from the data dump) that only had 1 tag. This helps it recognize particulars of a specific tag. Code and prose were split into separate features, as one of them may have greater impact on the result. When given a question, it returns the top 10 tag suggestions; actual tags are included for comparison.
Structure was a CSV file:
"tag","code blocks(200 chars)","body text(200 chars)"
Some characters were filtered out of code
that caused training errors: []^|~
. Not sure which of those was causing problems.
If you'd like to see a certain question added to the test list, let me know. Obviously, questions that have code predict better.
The other project predicts movie ratings based on IMDB data and director/actors. Unlike the tag suggester, this one is live so you can experiment with different combinations to see what it would predict.
Structure here was:
rating,"directorId","actorId actorId actorId"
Both are running on Google App Engine so python is the back end. I'm not using a particular API; just followed some of Nick Johnson's example code.