Quote Originally Posted by Dada View Post
I don't think you really appreciate the astronomical amounts of data needed to make any relevant predictions. Each team makes 8 choices (3 bans, 5 picks) from a pool of ~95 champions. Thats over 10^31 different possible combinations just there*.

Now, you need a fairly large sample for each of these combinations. 100 or even 1000 games is definately on the low end for statistical significance, but might suffice if we don't want to get our results published or anything.

Then there's the variance - runes, masteries and player skill, as well as play style. All the results would be heavily influenced by these "hidden variables", effectively making the data useless, unless you start considering these choices in your model as well. In that case, just the number of combinations easily exceed the number of atoms in the observable universe.

You don't have to discuss this if you don't want to, since you don't find the statistical aspec interesting, but I wanted to point out that, no, the data definately isn't there.

For the actual usefulness of either this tool or tier lists, I agree with the thread - the human factor has much to big of an effect, and you need to make the decisions yourself instead.

*If you count team1: Ban Shaco, team2: Ban Shen and team1: Ban Shen, team2: Ban Shaco as different outcomes.
Oh, the data is definitely there, and it's definitely possible to assemble it and do the statistics. It's just not something to do in a mornings work - but then I never said it was.

Of course, there is some leeway in the term 'feasible'. It could be done, certainly, if anyone really wanted to. Actually, with some distributed computing (as in, each players calculations are done locally and only the results uploaded to a central database) the astronomical number of calculations becomes much, much less of an obstacle.

But then again, the actual doing of the thing was never my point. It's like ... space elevators.