cheese: Bleh, I lost interest in GSC a while ago. The games are just too looooong.
Sept 29, 2015 6:26:12 GMT -8
magic9mushroom: Ugh. I'm doing a calc (specifics will be a SURPRISE), but one scenario that I thought would be relatively short has turned out to be absolutely horrific. It's actually rather important, too, so I can't just skip it.
Sept 30, 2015 22:12:58 GMT -8
magic9mushroom: The scenario has 12 outcomes, and most of them have half a dozen or more sub-cases each of which is a product of about a dozen probabilities. Holy combinatorial explosion, Batman!
Sept 30, 2015 22:20:54 GMT -8
piexplode: so what about 12x6x6? that's ~500 as an estimate then. At least that's still within plausible calculation (3 times as big as the lapras chansey calc?)
Oct 1, 2015 13:17:59 GMT -8
magic9mushroom: Well, not exactly comparable. Lapras/Chansey was probably bigger in an absolute sense than this scenario (because there were probabilities being multiplied in each of the states) but this one's more finicky.
Oct 1, 2015 18:07:42 GMT -8
magic9mushroom: Also, this is one scenario (which I initially thought would be like 10 lines of calculation) out of... well, it's the twelfth so far, and I think I'm about halfway through the problem (there's at least one even uglier scenario to come).
Oct 1, 2015 18:10:14 GMT -8
piexplode: So it's more like, lapras/chansey was easier to automate, this one has less work involved in terms of calculations required to solve, but it's harder to automate meaning that in a practical sense it involves more work, I guess?
Oct 2, 2015 10:32:15 GMT -8
magic9mushroom: The thing I'm complaining about is a matchup that's usually over in 1 or 2 turns. But both sides' moves have secondary effects, I'm including 255s, and there's a couple of cases where damage rolls matter.
Oct 2, 2015 19:20:08 GMT -8
piexplode: oh my. Maybe do one neglecting 255s first i guess, or are you wanting to be 100% accurate?
Oct 3, 2015 14:57:21 GMT -8
magic9mushroom: Well, part of the reason I'm doing this is because you neglected them. Also, given the ambitious nature of this particular calc, and the consequent likelihood that it will be linked to a lot, I'd like to only have one version of it.
Oct 3, 2015 16:05:42 GMT -8
piexplode: ah. Looking forward to seeing your results
Oct 4, 2015 1:32:29 GMT -8
piexplode: I wonder what other distributions, functions etc. you could put into competitive pokemon if you were to design your own set of pokemon including mechanics.. maybe you could have added effects work like say lambda = 1 for the effect: drop opponent's speed b
Oct 7, 2015 6:26:38 GMT -8
piexplode: ... by one stage, with that being a rate of 1 over a length of 5 turns (so I guess lambda = 0.2 for any one attack...) and I wonder if that'd be easy enough to code...
Oct 7, 2015 6:27:08 GMT -8
piexplode: http://pastebin.com/xTJ04zvF had some ideas and chatted with kristoph and blue kirby about them..
Oct 7, 2015 6:56:27 GMT -8
magic9mushroom: Uh-oh. I made a mistake somewhere in 31,000 characters of calculations, and now I have to find it.
Oct 7, 2015 17:33:49 GMT -8
magic9mushroom: Phew. Found both typos. I'd written (255/256 * 1/256 * 193/256 * 255/256) instead of (255/256 + 1/256 * 193/256 * 255/256) in two of my denominators.
Oct 7, 2015 18:01:00 GMT -8
magic9mushroom: I caught it because I always sum my probabilities at the end of a scenario and they summed to 1.0000003 instead of 1.
Oct 7, 2015 18:04:02 GMT -8
piexplode: at least you caught it!
Oct 7, 2015 23:06:36 GMT -8
magic9mushroom: I don't think it would actually have affected the results I was quoting, since I was only quoting to 4 d.p., but it's the principle of the thing. They're supposed to be accurate to 11 d.p. and they will be.
Oct 8, 2015 2:09:49 GMT -8