What part would you like me to respond to? I can hardly dump three classes worth of information and 8 years of experience in defense of my opinion on the topic.insertnamehere wrote: Well, since you have decided only to respond to that part, and not say how it actually will not work in your mind, but only automatically condemn me, is a bit suspect.
Of course, if you would like to release a list of the libraries you're using and at least some of your source code, I'd be happy to look for bugs for you and try it on some of my company's sample data sets to see how well it does classifying customer demographics or twitter spam. You might have some chance of success at that latter one, if your code is correct.
I actually did say how it will 'not work in my mind', though. It will not work because you will have insufficient data.
I have no idea why you think you will have enough. You'd need hundreds of games worth of info from a single
person to make any kind of assessment about them with regards to any concrete variable with even a simple Bayesian filter - I doubt there is a single
person playing who has played more than 50 games of online mafia. And as for your assertion that *not* having a single variable means that it's less of a problem... that is completely backwards. having two variables instead of one would *square* the required sample size for any given confidence interval.
And *that* all assumes there even *is* a concrete variable in there representing 'writing style while bad'. Which there of course is not.