[02:18] *** moon-child left [02:23] *** moon-child joined [07:09] *** patrickb joined [10:11] *** holly__ joined [10:12] Now there's a nice integral of squared-error loss in Game::Numeric, it can be used to predict random variables (see Game-Decision) [10:12] Uploaded today [10:13] Now I need to debug both and more or less add tests [10:14] There's just simple tests now in Numeric and Decision [10:14] So I have 3 standard loss functions [10:15] S/So/So now [10:16] S/So/So now [10:16] *barf* stupid irc client [10:17] holly__: I wanna read your stuff but there's a lot. What's the sort of ... entrypoint module in all of this? :D [10:18] Game::Stats is very basic statistics, then Game-Numeric is standalone numerical functions (integrals for decision theory) [10:19] Then Game::Bayes is bayesian statistics extended and Game::Decision is what you have to read last, decision theory based on the 3 other packages [10:20] There's some design faults as I forgot some things in Game::Bayes [10:20] *blush* [10:20] Game-Stats contains discrete distribution functionality [10:21] So a Game::Stats::DistributionPopulation contains just a list of chances [10:21] This is basically needed for calculating estimates, variances etc. [10:22] This is called a distribution in statistics [10:22] (Discrete) [10:23] Game-Bayes needs more work so you just have some brute MAP learning and that's it for now, all based on the disrtibutionpopulation of Game-Stats ^^ [10:23] Then Numeric speaks for itself, definite integrals for the bayesian analysis [10:24] Then decision theory is quite nice in Game::Decision as you have several loss and risk functions which can be used in "Game"'s [10:24] Decision theory is loss and risk functionality [10:25] So you get fractions/probabilities with rewards and so on to calculate potentials (with integrals) [10:25] These things are needed for adaptive systems [10:26] To be more informative : Game-Decision is the best to learn in the end as it has clear game functionality, don't mind the dependencies as they are simple [10:27] Now TBH it's just a main pattern class with math function all the time, for speed instead of OOP [10:27] These things are faster than non-core/core dispatches [10:27] HTH [10:28] I see I see [10:29] I forgot I asked a question in IRC but I appreciate that you answered thoroughly anyway :D [10:29] Means I can digest it when I can manage [10:29] Sure, take care [10:37] In Game::Decision is an algorithm for 3 rewards r1 << r3 << r2, << == preferred over [10:38] That's also usable in games then there's just loss functions [10:38] These 2 are the basics of the whole system [10:38] Risk functions will come later together with Bayesian analysis which is for e.g. inference [10:39] * holly__ needs to finish his book chapter [10:39] (About non-informative priors now) [10:43] These things will need more time as I have to understand the CS features of the noninformative priors [10:43] I need Haar measures and Fischer info matrices [10:44] Which probably will bootstrap Game::Matrix etc. [10:46] To conclude, if you understand the functionality of Game-Decision you'll be ok, the 3 dependencies are to be comprehended but are more easy [10:47] And for games pretty simple to use as they are just statistics and small functionality [11:01] *** holly__ left [11:27] *** holly__ joined [11:38] I am going to stop working on my Game:: libraries for this week [11:40] I'll rejoin on sunday [11:40] Evrything should be online in 6-12 hours [11:40] bye! [11:40] *** holly__ left [11:48] *** holly__ joined [11:49] I'm going to stop working on my Game::X libraries, everything should be online in 6-12-24 hours. So everyone has it for the weekend. CUL [11:49] (for this week) [11:49] *** holly__ left [14:54] *** lizmat_ joined [14:57] *** lizmat left [16:31] *** GreaseMonkey left [16:50] *** greaser|q joined [18:15] *** holly_ joined [19:57] *** holly_ left [21:44] *** lizmat_ is now known as lizmat [22:05] *** patrickb left [22:58] *** japhb joined