It's not quite clear to me what this actually does. How does training work? Do I give it input and actual decisions so that it can decide based on trends or is it more of a fuzzy match to known data?
It has been a while since I have checked it out, but you give it training data (examples + values) and then it will be able to predict values for unseen examples. Eg. predict if a comment on a blog post is spam.
Actually the google prediction api is already really old; I don't know why it shows up now on HN
