@vzn wow, what a good news, thank you ;)

Hey guys!

I'm just starting out with machine learning, and i've got a gigantic database of economy transactions on my roleplay gameserver

There's various groups on my server. Like there's drug dealers, weapon dealers, cops, firemen, mafias e.t.c. I was wondering if it'd be a good learning project to try to cluster various transactions?

The only data that i have would be a transaction log transferring $X from some player A to B, nothing more. I was wondering for example that if i knew some of the mafia members on the server, would it be a good ML learning project to try to find all the other mafia members depending on who they interact with? Or would that be too advanced for a total beginner?

I can also collect pretty much any other data on the server. In particular i had the idea of tracking each player's coordinates and finding the closest players to them to try and figure out who they hang out with and so on

@AayushAgrawal hi, welcome to room, sounds like interesting prj! is this for school or personal? do you have CS/ math bkg? plz summarize. yes a lot of game data is ripe for a lot of big data analysis. there are many angles. it depends on your goals. need to understand your app better to make recommendations. suggest you survey scientific papers wrt the game type youre using.

@Evil yw, but why "good news"?

@vzn because I do this kind of experiments for a long time now and instead of trying to convince anyone that the idea is not stupid and the results are not magic or coin toss, now I can show that now this is a "news".

@Evil :)

@vzn even though there are several percents of cases where it must fail and I do not accept the argument of majority (here simply pointing fingers at similar research) the majority of people does ;)

@vzn I am currently occupied with salvaging some exotic and faulty device. I am not sure why anyone would make sampling rate low and odd using one electrode. One bump, high frequency, rapid move... and the signal is lost.

@AayushAgrawal Sounds like a potentially fun project to me! But only you can tell whether you'd find it fun. Your ability to experiment with it might rest partly on whether you have labelled data (i.e., you have ground truth for some people -- you have a set of people where you know what types they are).

As far as how challenging it will be, that's probably very dependent on the specifics of how people tend to behave in that particular game and how distinct the behavior of each type/group is. You can always try something simple (just a few features, a very simple ML algorithm) and see how well it works. That's one nice thing about machine learning.