|
United Kingdom20168 Posts
it is anti proportional
-10% time = 11% faster -33.33% time = 50% faster -50% time = 100% faster -90% time = 900% faster -98% time = 4900% faster
etcetc. Each percentage of time reduction creates a greater speedup as it's a higher percentage of the remaining time - i'm not sure that exponential describes this properly, but i dont think inversely proportional directly and fully describes it either
|
On July 21 2016 18:30 Laserist wrote: I got bored of my previous land locked run and decided to start a new game with harder than normal difficulty, and aggressive AI. And my ass is handed to me by the nearest neighbor. GG wp, will play against cheater AI for sure 11/10. Coding a semi-competent AI is that hard?
Ehm yes, yes it is. Practically every harder AI cheats because it turns out that writing the equivalent of Deep Blue is very hard.
|
On July 21 2016 18:30 Laserist wrote: I got bored of my previous land locked run and decided to start a new game with harder than normal difficulty, and aggressive AI. And my ass is handed to me by the nearest neighbor. GG wp, will play against cheater AI for sure 11/10. Coding a semi-competent AI is that hard?
How many (strategy-) games do you know that manage it? Yes, it's very hard.
|
On July 21 2016 19:36 Gorsameth wrote:Show nested quote +On July 21 2016 18:30 Laserist wrote: I got bored of my previous land locked run and decided to start a new game with harder than normal difficulty, and aggressive AI. And my ass is handed to me by the nearest neighbor. GG wp, will play against cheater AI for sure 11/10. Coding a semi-competent AI is that hard?
Ehm yes, yes it is. Practically every harder AI cheats because it turns out that writing the equivalent of Deep Blue is very hard.
Deep Blue was a brute forcer too. IIRC the "best" Starcraft AI just abused muta micro or something like that with minimal improvements to most stuff and a few scripts to enable better building placement and resource management.
|
There are some games utilizes a more advanced AI, like Grey Goo (learns player moves and try to counter them the more you play). I don't have enough experience to justify how good they are. I was not asking a deep blue rather, slightly more intelligent than what we have now.
|
On July 21 2016 22:42 ticklishmusic wrote:Show nested quote +On July 21 2016 19:36 Gorsameth wrote:On July 21 2016 18:30 Laserist wrote: I got bored of my previous land locked run and decided to start a new game with harder than normal difficulty, and aggressive AI. And my ass is handed to me by the nearest neighbor. GG wp, will play against cheater AI for sure 11/10. Coding a semi-competent AI is that hard?
Ehm yes, yes it is. Practically every harder AI cheats because it turns out that writing the equivalent of Deep Blue is very hard. Deep Blue was a brute forcer too. IIRC the "best" Starcraft AI just abused muta micro or something like that with minimal improvements to most stuff and a few scripts to enable better building placement and resource management. It will always brute force. By the time you have an AI able to use intuition you are well beyond a game opponent and into the realm of creating a digital person.
These games are so incredibly complex with so many decision points at all times. They are utterly beyond our ability to create a truly challenging AI without it resorting to cheating.
|
On July 21 2016 22:49 Laserist wrote: There are some games utilizes a more advanced AI, like Grey Goo (learns player moves and try to counter them the more you play). I don't have enough experience to justify how good they are. I was not asking a deep blue rather, slightly more intelligent than what we have now.
uh that would be scary in stellaris. Just imagine every empire near you would counter your usual research and ship design route. Even normal AI would stomp you. But its hard to make something like that work when the start is almost pure rng. It works well in a rts that doesn't have many starting routes though.
|
On July 21 2016 22:57 Gorsameth wrote:Show nested quote +On July 21 2016 22:42 ticklishmusic wrote:On July 21 2016 19:36 Gorsameth wrote:On July 21 2016 18:30 Laserist wrote: I got bored of my previous land locked run and decided to start a new game with harder than normal difficulty, and aggressive AI. And my ass is handed to me by the nearest neighbor. GG wp, will play against cheater AI for sure 11/10. Coding a semi-competent AI is that hard?
Ehm yes, yes it is. Practically every harder AI cheats because it turns out that writing the equivalent of Deep Blue is very hard. Deep Blue was a brute forcer too. IIRC the "best" Starcraft AI just abused muta micro or something like that with minimal improvements to most stuff and a few scripts to enable better building placement and resource management. It will always brute force. By the time you have an AI able to use intuition you are well beyond a game opponent and into the realm of creating a digital person. These games are so incredibly complex with so many decision points at all times. They are utterly beyond our ability to create a truly challenging AI without it resorting to cheating.
Agreed. Google DeepMind used neural networks instead of brute force to figure out Go which was pretty impressive, but Go is a game with pretty simple rules.
On July 21 2016 23:19 FeyFey wrote:Show nested quote +On July 21 2016 22:49 Laserist wrote: There are some games utilizes a more advanced AI, like Grey Goo (learns player moves and try to counter them the more you play). I don't have enough experience to justify how good they are. I was not asking a deep blue rather, slightly more intelligent than what we have now. uh that would be scary in stellaris. Just imagine every empire near you would counter your usual research and ship design route. Even normal AI would stomp you. But its hard to make something like that work when the start is almost pure rng. It works well in a rts that doesn't have many starting routes though.
How do they counter my tach lance deathball?
...oh. More tach lances than me b/c AI can macro better than me :/
|
On July 21 2016 23:19 FeyFey wrote:Show nested quote +On July 21 2016 22:49 Laserist wrote: There are some games utilizes a more advanced AI, like Grey Goo (learns player moves and try to counter them the more you play). I don't have enough experience to justify how good they are. I was not asking a deep blue rather, slightly more intelligent than what we have now. uh that would be scary in stellaris. Just imagine every empire near you would counter your usual research and ship design route. Even normal AI would stomp you. But its hard to make something like that work when the start is almost pure rng. It works well in a rts that doesn't have many starting routes though.
It doesn't actually track your in game movements and does counter moves like map hackers. It is supposed to analyze the way you play and react better next time. Basically it trains itself with you.
Google did a similar thing with Go and dump all known Go plays to an AI and let it play with itself again and again and again. Similar but much much more advanced.
|
On July 21 2016 18:36 Cyro wrote:-10% time = 11% faster -33.33% time = 50% faster -50% time = 100% faster -90% time = 900% faster -98% time = 4900% faster etcetc. Each percentage of time reduction creates a greater speedup as it's a higher percentage of the remaining time - i'm not sure that exponential describes this properly, but i dont think inversely proportional directly and fully describes it either
Ah. I was comparing the remaining percentage to the amount of research done in the same time
(100-10)%= 90% ---> 111% Research done 66.6% --> 150% Research done 50% ---> 200% Research done 10% ---> 1000% Research done.
And that is indeed an anti-proportional relationship. (f(x) = k/x)
For your numbers, the graph shifts a bit, since you basically replace "x" with "1-x", which means that you mirror the graph at the y axis and shift it by 1 (or 100%) to the right on the x axis.
This does not change the type of behaviour that the graph display. It is no longer Anti-proportional, as that name only applies to the k/x function. But you still have a pole at 100%, and your graph moves towards infinity as you approach that pole.
Exponential describes a different type of behaviour, namely that of the function f(x) = b*a^x with b any number and a any positive number. This function does not have any poles, but it does grow rather rapidly when approaching large numbers. Notable of exponential growth is that for a fixed change of the x value (time, for example), the amount of stuff is MULTIPLIED by a set number. So after one day, you have 1 apple, after two days you have 3, after 3 days 9, after 4 days 27 and so on. That is exponential growths.
Sadly, a lot of people think "exponential" means "growths a lot".
On July 21 2016 22:42 ticklishmusic wrote:Show nested quote +On July 21 2016 19:36 Gorsameth wrote:On July 21 2016 18:30 Laserist wrote: I got bored of my previous land locked run and decided to start a new game with harder than normal difficulty, and aggressive AI. And my ass is handed to me by the nearest neighbor. GG wp, will play against cheater AI for sure 11/10. Coding a semi-competent AI is that hard?
Ehm yes, yes it is. Practically every harder AI cheats because it turns out that writing the equivalent of Deep Blue is very hard. Deep Blue was a brute forcer too. IIRC the "best" Starcraft AI just abused muta micro or something like that with minimal improvements to most stuff and a few scripts to enable better building placement and resource management.
For Star Craft AIs, take a look at SSCAIT (Student Star Craft AI Tournament). I find it incredibly fascinating, and some of those AIs play in a rather cool fashion. Sadly, all of them are currently at most at D-level. Usually they have some major flaws in decisionmaking that hold them back a lot. Apparently, making even basic decisions in Starcraft is something that is really hard for AIs.
|
United Kingdom20168 Posts
Just got hit with the same bug for the 4'th time this game (an AI faction splits itself into a bunch of tiny factions and then declares war many many times on another)
it's making game crash sometimes when they instantly declare war many times (last one was 102 wars in the same game tick), is this a common bug?
-----
played another 8 hours today - long enough to regret ever playing on a large galaxy. Holy shit. I'm on a 6700k @ overclock (almost certainly the fastest cpu for running the game) and my friend is hosting on a stock 4770k, we've seen the game slow to 1/5'th of normal speed just from flying across a system with 300k worth of fleet. It's not worth it.
|
afaik u need fast ram i had 16gb ddr4 2666mhz and the game was running at fast speed lategame
|
United Kingdom20168 Posts
I have 3200 ddr4 from skylake launch
Prethoryn seem to like flying around with 2000 individual transport ships which kills performance, there were no major issues until that happened. I'l have to try it without somebody hosting on a weaker CPU. It doesn't seem easy to find any decent benchmarks for this engine either~
|
Never had that happen.. But so far I never had a dictator die without an heir, which should trigger this splitting event.
|
United Kingdom20168 Posts
|
United Kingdom20168 Posts
People still playing? :D
The plantoids cosmetic dlc (new portraits and ship/station models) just released to some controversy about pricing (at $8 with a dev saying that he thought it would have been worth it at $20)
There is a popular unofficial bug fix mod - a few of these things just got fixed in the plantoids patch but most are still bugged. http://steamcommunity.com/sharedfiles/filedetails/?id=728486525&
|
It seems that paradox went all in milking the customers.
https://reddit.com/r/paradoxplaza/comments/4w4e5p/stellaris_plantoids_species_pack_now_released/d63yphs
No, it's the sign they're sure that players will be ok with this. Paradox had record sales with Stellaris and HoI4, they are not struggling. The sign of them struggling would be fixing their games and making DLC system more comprehensive instead of asking players to spend 200 bucks on a game.
*****
I'm afraid that HOI4 has some mechanics intentionally broken to fix them via DLCs
|
You do know you just made general statements and accusations without argument or logic attached? Thats not how forums work at all. How is anyone suppose to respond to anything you said positively.
|
TAIWAN NUMBAH WAN5954 Posts
well the price for that pack is really hefty :/ but the claims about HoI4 being mechanics intentionally broken is just bullshit. It's the usual Paradox release, but I agree on HoI4 not living up to expectations
|
On July 21 2016 22:57 Gorsameth wrote:Show nested quote +On July 21 2016 22:42 ticklishmusic wrote:On July 21 2016 19:36 Gorsameth wrote:On July 21 2016 18:30 Laserist wrote: I got bored of my previous land locked run and decided to start a new game with harder than normal difficulty, and aggressive AI. And my ass is handed to me by the nearest neighbor. GG wp, will play against cheater AI for sure 11/10. Coding a semi-competent AI is that hard?
Ehm yes, yes it is. Practically every harder AI cheats because it turns out that writing the equivalent of Deep Blue is very hard. Deep Blue was a brute forcer too. IIRC the "best" Starcraft AI just abused muta micro or something like that with minimal improvements to most stuff and a few scripts to enable better building placement and resource management. It will always brute force. By the time you have an AI able to use intuition you are well beyond a game opponent and into the realm of creating a digital person. These games are so incredibly complex with so many decision points at all times. They are utterly beyond our ability to create a truly challenging AI without it resorting to cheating.
However modern chess engines are not bruteforcing. They are much more sophisticated and use alrgotims and decision tables to drop deadbranches (which Deep Blue would search and evalaute). So improments are defiently possible. The thing is both SC and Paradox games are much more complicated and would need a lot of additional work and also some decision process to evalute values of information not currently available to engine (like hidden in fog of war).
There is a lot of work to be done, but good Ai can be achived without coming even close to digital person.
|
|
|
|