Discussion: Ideas about evolution of players.

pier4r
Skynet
Skynet
Posts: 3390

Discussion: Ideas about evolution of players.

Post#1 » 18 Mar 2017, 10:45

Notes: AIs, tactics, programs are used mostly interchangeably in this post.

Some weeks ago with nullpointer on the telegram chat we had a discussion (at the end we disagreed) about the following point: I was insisting that a stable top3/5 player, playing after a week (or more) of pause and without changes, would lose most of the points falling in top10 or top20. This under the condition that the game mechanics would not change much. Nullpointer did not see it in this way.

Now in this period I did not play much, neither I modified my AIs during pauses, still I was impressed that when I played from time to time, meeting top20 players, I was able to hold the ground instead of just losing. Checking other players seems to confirm that, for example Tralalo did not play much between February and March and still holds his spot every time he plays again.

Of course after a while one changes little stuff, but the point is that "the first impact after a long pause is not devastating".

So, it seems that nullpointer was more near to the truth for the observations made.

Still I have the below ideas about evolution of player strength.

Given a range for scores (1), say 1000 as the first score and 2000 as the maximum score, and the current tiered leagues (there are 7) with XP levels (there are 7) to unlock bot classes and AI controls. Assuming that the top 2 players are around 2000 points, top 5 around 1900, top 10 around 1800, top 20 around 1700. I would say the following.

- from match 1 to 150 one is still discovering the basic qualities and weaknesses of bot classes and basic functions of AI controls.
- after match 150
-- up to 1400 score points: the player has still to work out the basic structure of his AI (whatever is a generalized AI, a generalized AI per bot class or specialized per map and bot class). This means: the player did not yet define in a stable way how his bots should behave in attack, retreat, resource capture, and so on. From time to time the player makes quite big changes for certain behaviors.
-- 1400-1600 points: the player has a basic structure, but now it starts to be complicated because games against stronger or equal players are now decided by smaller differences, or optimizations. This means that the general behavior from the AI when the player was 1400 is not so different, but some small changes in nodes, filters and priorities helps a lot in terms of results.
-- 1600-1700 points: the player starts to have the first useful subtle optimizations to hold his ground. So, once again, the basic behavior of the bot is there since the player left 1400 points, but now it is more refined to handle specific situations instead of the generic case.
-- 1700-1800 points: even more optimizations and this is the level, in my idea, that can consistently snatch points from the top10.
-- 1800-2000 points: way more optimizations than before, but, and this is my idea that so far did not hold the observations, I thought that the continuous competition between players 1700+ would produce more and more optimizations for particular cases and/or increasing efficiency in scoring balls or pushing or damaging the enemy. This means, in my idea, that a top 10 player not playing for a week, would be at least 300 games of experience behind the current optimizations. Therefore losing a lot of ground before catching up and going back to his previous spot.

End of the exposition of my ideas about evolution of player (if it is not clear, ask!).

Anyway, this seems not valid. At least for the check that I did, through the stats page or personal experience, it seems that top 10 players not playing for more than a week (and likely not editing their AI during the pause) just hold the ground without losing a lot of points. Surely they lose a bit more than before, but still not horribly, before catching up.

Why is it so? I do not even have partial answers at the moment. What do you think about it?

(1) If one does not set a range, scores are meaningless because while the starting score is still the same as in November 2016 (1000), the top scores are increasing.
See
- viewtopic.php?f=22&t=126&p=3880#p3880
- viewtopic.php?f=7&t=414
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

Botique
Algorithm
Algorithm
Posts: 96

Re: Discussion: Ideas about evolution of players.

Post#2 » 18 Mar 2017, 15:15

I think (cant proove) that the game is still somewhat sime and players are finding some local optimum on the bot AI landsape. And it seems game itself leads most of top playera around the same "peak". Findig other peak would require some real disoveries but coz not much meta changes the player after a week are still around the same "mountain" of AI optimization.
What you expect would change if there would be paradigm shift... some qualitative changes in meta... but it does not happen and I think its coz of game complexity.
But even slight change in game rules or intoducing something new might drastically change the game meta anf then it might be that some players would beat the tops.

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#3 » 18 Mar 2017, 16:51

Of course my idea is valid with the same version, so if the game changes even a little everything is discussed again. But I like the idea of local optimum, as indeed should be. Yet for the score formula even being a tiny bit higher in the mountain(say, a couple more of optimizations) would mean a widening gap of scores.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

harthag
Automaton
Automaton
Posts: 197

Re: Discussion: Ideas about evolution of players.

Post#4 » 18 Mar 2017, 17:42

When I very first started lurking, before even joining the forum, I saw a post that said something like "there are a finite number of situations a bot can be in, so it should be possible for a single AI to be prepared to handle them all".

Given your restriction of "no chamges to the game" for the discussion, that concept should be true, though the actual implementation is probably only theoretical due to the sheer number of possible situations.

However, while the "complete" implementation might only be theoretical, I suspect that the top 3 or 5 players have reached a "practical" level of implementation, which allows them to hold their points and positions. In other words, their AIs have reached a point that they are "effectively" ready for "anything" the current game can produce. Or close enough that they are not significantly affected by lapses in play because there is nothing new (nothing their AIs are unprepared for) that others could come up with in the meantime.

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#5 » 18 Mar 2017, 18:23

@harthag,yes it would be more or less like Botique said, just I don't believe top players are so near to the very perfected optimizations. I feel there is an host of subtle stuff still not yet used otherwise it could not be that players continously make changes.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

mcompany
Autonomous Entity
Autonomous Entity
Posts: 872

Re: Discussion: Ideas about evolution of players.

Post#6 » 18 Mar 2017, 19:02

I will note that there has not been a significant change in the game and how a match could function since 5.3, when bot classes got limited, and because of this, we've had plenty of time for optimizing. And the closer we get to optimal, the longer it takes for a new optimization to be better than the last. I will like to mention that I have an account that still purely uses an AI that is nearly completely unchanged since 4.2, with the only changes being bug fixes. (http://gfx47.com/games/Gladiabots/Stats ... ay=matches). While being heavily unoptimized, it is still enough to get me into Master league, even though I made the original AI while I tried to figure out what I was doing. Likewise, during 5.2 and 5.3 nullpointer spent large amounts of time not playing and each time he came back, he was as strong as ever (and I've tried to stay up too before... A couple of losses and you can fall extremely quickly). Considering how optimized that current top players are, I do not think that staying away for a period under a month would put you that far behind unless a major discovery was found (similar to how the rush tactic on Mind Game is currently one that I don't think people have found a counter to, so most of the players has switched to uses some sort of shotgun/machine gun combo instead of sniper-machine gun)

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#7 » 18 Mar 2017, 19:19

mcompany wrote: And the closer we get to optimal, the longer it takes for a new optimization to be better than the last.


I like this point. It seems quite reasonable and explains, together with other points, why my expectations are not met so far. Because the top players either are new and are rediscovering already known optimizations or are old players that needs still to find out the next round of sensible optimizations.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TAI
Hello World
Hello World
Posts: 8

Re: Discussion: Ideas about evolution of players.

Post#8 » 18 Mar 2017, 21:39

First of all, evolution implies a tree branching out, not a ladder with one direction.

I think the game is actually closer to a complex form of rock-paper-scissors. There isn't really one perfect way to optimise your bots. As some of you mentioned earlier, while there is a finite number of situations a bot can be in, choices are also limited. Choosing a bot's behaviour in a specific situation possibly excludes other options in that same situation.

The map Mind Game is probably the best example of this. Although the shotgun/machine gun combo is indeed a mighty rock, it doesn't mean the paper doesn't exist.

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#9 » 18 Mar 2017, 22:28

Yeah but every player has programs a bit different so that for me accounts for evolution. It is not that we end up with the same behavior more or less. Let's say that is very close evolution.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#10 » 19 Mar 2017, 18:56

Update: meeting a lot of different 1700+ players today, and a lot of points are burned, so my idea may be partially wrong but maybe takes a bit of time -namely one should expose the weaknesses - to be caught up..
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#11 » 19 Mar 2017, 22:40

From the telegram chat.

mcompany wrote:I mean that when it comes to Grand Master, 1300-1500 is generally the players that either should be in Master but they barely got a high enough score or they are players who haven't got to level 6 yet, but they are high risers, 1600-1700 are the players that can hold their own, they win sometimes, but they aren't particularly strong, 1700-2000 seems to be the strong players who are fighting for being the best, but they always have some map or bug that consistently beat them, and over 2000 seems to be players that are just REALLY strong. I don't think I've seen a single player fall downwards through these zones, and often, the players get stuck in whatever that respective zone is for that period of time.

Now what I'm also saying is that from 1700-2000, player continuously go from the bottom to the top of the zone and back again almost daily. Likewise, a seasoned player that could win nearly every tournament would go from 1950 to 1780 from maybe 5 loses and then get back to 1950 with very little changes to their AI by a whole day full of mostly wins
People in 1700-2000 are just so unoptimized that one loss could destroy everything and rearrange the leaderboard, but it still wouldn't mean anything)
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#12 » 26 Mar 2017, 09:26

So tct played after well one month and, while I guess he fixed something since the last losses in February, he just keep winning 6 matches.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#13 » 26 Mar 2017, 19:34

pier4r wrote:So tct played after well one month and, while I guess he fixed something since the last losses in February, he just keep winning 6 matches.


But after that he lost quickly a couple, saying also in telegram that the others improved a lot. So, part of me still clings on the idea that I exposed.
Attachments
2017-03-26 19_32_58-Mozilla Firefox.png
2017-03-26 19_32_58-Mozilla Firefox.png (99.55 KiB) Viewed 2504 times
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#14 » 07 Apr 2017, 09:00

Another things that I cannot explain at the moment is why the league immediately before the top league is always smaller than all the others (excluding silver and bronze).

It happened also before creating grand master. It is like the league before the top league is just used as transition. Then one either goes two leagues before the top league or stays in the top league. Somehow players do not stay too much in the master league (and before February, diamond league).

But I did not pinpoint the real explanation.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

User avatar
Bombafat
Script
Script
Posts: 40

Re: Discussion: Ideas about evolution of players.

Post#15 » 07 Apr 2017, 14:21

pier4r wrote:Another things that I cannot explain at the moment is why the league immediately before the top league is always smaller than all the others (excluding silver and bronze).

It happened also before creating grand master. It is like the league before the top league is just used as transition. Then one either goes two leagues before the top league or stays in the top league. Somehow players do not stay too much in the master league (and before February, diamond league).

But I did not pinpoint the real explanation.


I think it has something to do with the fact that grand masters has no cap. So normally we should expect the highest amount of player at the "average" league. As higher we would go: Platin-->Diamond-->Masters, the less players are to be expected. But once reaching Grand Masters, they reached the highest possible league. Since the highest league is not restricted to a certain percentage of players or a limited number (like only best 100), more and more players can eventually reach Grand Masters. Therefore the "law" that the number of players decreases as higher we go in league holds not true anymore for Grand Masters, therefore the league below will always have the lowest number. Maybe i am wrong but this is my idea about it.

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#16 » 07 Apr 2017, 22:32

Hmm, it makes sense.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#17 » 03 May 2017, 22:14

So after Tralalo and TcT (Tct playing less than tralalo though) playing after several weeks of pause, today T-800 appeared again after more than one month. Now I assume nullpointer deployed matches without fixing his old AI immediately (I assume the same for tralalo and TcT). T-800 got some defeats but then those were apparently fixed quickly. So once again it seems that while old "good" AIs may be not up to date, it does not take much to fix them, if the game mechanics are the same.

What I'm saying is not valid if nullpointer (and Tralalo and TcT) worked on his ai (and studying replays of the current best players) before deploying it after months.
Attachments
2017-05-03 22_18_07-Mozilla Firefox.png
2017-05-03 22_18_07-Mozilla Firefox.png (100.79 KiB) Viewed 2389 times
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#18 » 04 May 2017, 10:19

So yesterday T800 played 80 games and lost 24. This I would say that is a better evidence that strong "old" tactics (if the game does not change) are still competitive but not anymore as efficient as before. Still they should hold the top 10 if the amount of active players is the same. For example they "degrade" from 5% defeats only to 30% defeats.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

harthag
Automaton
Automaton
Posts: 197

Re: Discussion: Ideas about evolution of players.

Post#19 » 04 May 2017, 14:34

pier4r wrote:... evidence that strong "old" tactics (if the game does not change) are still competitive but not anymore as efficient as before...

Is this a good thing? Bad thing? Neutral observation? What does it mean for the future of the game?

pier4r
Skynet
Skynet
Posts: 3390

Re: Discussion: Ideas about evolution of players.

Post#20 » 04 May 2017, 14:41

I Don't know. I can just say what I suspect since long time : that the game is far solved so even new tiny discoveries can bring a better optimum. Those discoveries need several weeks to be found since after a certain skill it is not so trivial to see improvements. I think this allow dominant players to be top 10 after months of inactivity, because the evolution of active players happens but not so fast.

I would expect that after one year without changes to the game, then a previous top player would drop the top 10 without recovering quickly.

Of course this is even more valid if the game mechanics change and expand.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

Return to “General Discussions”

Who is online

Users browsing this forum: No registered users and 2 guests