Grand master league?

pier4r
Skynet
Skynet
Posts: 3390

Re: Grand master league?

Post#21 » 05 Feb 2017, 00:53

And once again: team tags so far accomplish a little (mcompany is tagless for example). The sniper is important.

But yes, in whatever competition, thrown against someone too experienced is not giving much insights. I mean I have difficulties to understand the mistake against top players too.

It is like someone just learned to say "hello" in a language and the other expect that this poor guy knows his language 100%.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

User avatar
Ritter Runkel
Neural Network
Neural Network
Posts: 498

Re: Grand master league?

Post#22 » 05 Feb 2017, 00:58

pier4r wrote:I mean I have difficulties to understand the mistake against top players too.


Me too. In Grand Master's league I understand maybe 30% of the matches I lose. So mostly I see that I lose but I've no clue why.

Luckily the 30% (when improved) fixes bugs I didn't saw in other matches

User avatar
GFX47
Dev
Dev
Posts: 2914

Re: Grand master league?

Post#23 » 05 Feb 2017, 09:35

Shmalzbrot wrote: How about displaying the current elo score instead of just “(top league)“? Is that worth a feature request?


Yep ;)
https://trello.com/c/18jlS2DC/465-displ ... vel-league

Botique
Algorithm
Algorithm
Posts: 96

Re: Grand master league?

Post#24 » 06 Feb 2017, 09:49

Ohh... !$@% just got promoted, even as I didn't played for some days. Now I have to remember what I have programmed my bots, and I still didn't feel my bots are any good...
Now I have to come up with some clever programming... but already feel that I need something new. Old ways does not cut it anyway.

harthag
Automaton
Automaton
Posts: 197

Re: Grand master league?

Post#25 » 06 Feb 2017, 20:12

Not sure if this is the appropriate thread for this, but I haven't come across a better one for it yet (new to the forum, trying to catch up):

There is a definite issue with matching players for battles, and I admit I exploited it to get in to grand master league (I probably deserve Master, but not Grand). Certain players play more often at certain times, so it's possible to spam matches that you are more likely to win, if you spam them at certain times of day or certain days of the week, etc.

Possible solution: Allow offline players to be pitted against online players, using the most recent (or maybe the most recent victorious) AIs selected by the offline player in a previous battle, and at no risk to the offline player for a loss.

I'll use the most famous (as far as I can tell) top player as an example: If I'm online and Nullpointer is not, and I initiate a battle, give me some chance of being matched with Nullpointer's most recent AI set from a previous battle Nullpointer fought (and won?), If I win, I get the points, but Nullpointer loses nothing (or minimal loss?). If I lose, I suffer the consequences, but again, Nullpointer gets no benefit (or reduced benefit?).

Just a thought.

pier4r
Skynet
Skynet
Posts: 3390

Re: Grand master league?

Post#26 » 06 Feb 2017, 21:24

harthag, what you found is a possible problem (especially if you follow the stat page. I follow it to jump in when ritter, mcompany, tralalo and co are playing when I'm in tournament mood).

I would say that it gets solved either by more players, so randomness, or by the will to play. One can play around to get score there and there, but slowly there will be cases where one has to be matched with a strong opponent and if one is not prepared, is going to lose truckloads of points due to the score formula. So what you raise as problem can be applied not for long.

There are the following cases:
a. the player is very strong, so s/he mostly wins, therefore does not matter when to deploy matches.
b. the player is not so strong. Sooner or later he will either get paired with a strong opponent (if he wants to get more points, due to how the score formula works) or a stronger opponent raises because the opponent is new (see note 1) and will lose a lot of points from this player due to the score formula.
c. existing players improve. See b.

Therefore it will be a very tedious work to always avoid the ones that are better than you. It may work for the moment, if you are actually stronger than your opponents, but then not anymore.

And being in grand master is not so difficult. If you note from the stats page, grand master slowly accumulate people while the league "top-1" is shrinking. I still have to formalize why but I suppose that the possibility of interleague matches, allow the ones near to the demotion threshold to survive, because at the end they are stronger than those in the league below, so they get enough points. This is ok since they can improve in the meanwhile.

This may also complicate the situation because, unless players gets more or less equally competitive, this would mean a bigger and bigger gap between top20 and the rest. We will see.

note1: we have "generations" of players now. The ones from alpha 5.2 (jbdb, ritter, blitz, pier, conscious, nullpointer, mcompany, crashdown, lord valcor, peli, arcueid that had two accounts, etc) the ones from alpha 5.3.1/6.x (dirk, tralalo, counterbalance, kjr, dangeur, valerian, koza, kajikira, etc...), the ones from alpha 7.8 (tct, rock and roll, igigin, grindmax, titus, aw0093, and many others)
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

User avatar
Ritter Runkel
Neural Network
Neural Network
Posts: 498

Re: Grand master league?

Post#27 » 06 Feb 2017, 21:28

pier4r wrote:note1: we have "generations" of player now. The ones from alpha 5.2 (jbdb, ritter, blitz, pier, conscious, nullpointer, mcompany, crashdown, lord valcor, peli, arcueid that had two accounts, etc) the ones from alpha 5.3.1/6.x (dirk, tralalo, counterbalance, kjr, dangeur, valerian, koza, kajikira, etc...), the ones from alpha 7.8 (tct, rock and roll, igigin, grindmax, titus, aw0093, and many others)



And now I feel old :?

pier4r
Skynet
Skynet
Posts: 3390

Re: Grand master league?

Post#28 » 06 Feb 2017, 21:34

Well game wise, we are old. It is not so easy to get that amount of matches and XP (this means not only blind spam, but also development). One has to spend time in the game over days, weeks, etc.

In general the direction of the game is still pleasant compared from alpha 4.3. Just alpha 5.2 was a bit meh, because the bot choice and the placement was too important. I mean it is still important, but now it is subtle.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3390

Re: Grand master league?

Post#29 » 06 Feb 2017, 21:36

Anyway from alpha 4.x I rember peli, blitz and lord valcor (I played just ~200 games in alpha 4.3). Then I have no data anymore.

Also: http://gfx47.com/games/Gladiabots/Stats ... ofnormandy was a player in the top20.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

harthag
Automaton
Automaton
Posts: 197

Re: Grand master league?

Post#30 » 06 Feb 2017, 23:47

pier4r wrote:There are the following cases:
a. the player is very strong...
b. the player is not so strong. Sooner or later he will either get paired with a strong opponent (if he wants to get more points, due to how the score formula works) or a stronger opponent raises because the opponent is new (see note 1) and will lose a lot of points from this player due to the score formula.
c. existing players improve. See b.


I don't fit any of those, exactly, and that's my point. I'm definitely the "not so strong" you mentioned in "b", but I effectively avoided the strong opponents during the "sooner", and I kept it up long enough to get promoted "later". Sure, I met my fair share, and lost a lot of points along the way, but that's how I learned when they play. If my suggestion were implemented, I doubt I would ever have made it to Grand Master, I'm just not good enough (in my own eyes anyway).

pier4r
Skynet
Skynet
Posts: 3390

Re: Grand master league?

Post#31 » 06 Feb 2017, 23:56

Hmm, since at the end master league is where people really apply stuff, I think it is ok. The problem would be if it would be always possible, but it is unlikely for the moment . Yes one reaches master league but for the current player base whoever has a minimum ability in logic and analysis can do it. The others either do not try, or are confused or are not interested.

So welcome in League 7 and please release your points.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 565

Re: Grand master league?

Post#32 » 26 Mar 2017, 23:37

Hi guys.
I would know something i dont understand:
How people stay in grand master league with a % of win under 50%?
I mean if they lose more than they win how can no fall in the previous league?

pier4r
Skynet
Skynet
Posts: 3390

Re: Grand master league?

Post#33 » 27 Mar 2017, 00:26

You should get the idea how the elo score works.

See: viewtopic.php?f=22&t=126 (section score explanations)

the elo score works great but it is not so common in online games so many gamers are used to check the win ratio, that is a mediocre measure of strength (at least until someone is very good).

So in short: if there is a group of players that obliterates you, but you lose every time 1 point from them, and then every time that you win you win 16 points, you compensate 16 defeats with one win. (I see what you think "but this is unfair!" No, stop, first understand the score formula and its qualities, then complain)

Second part, but this could be solved with simple math from, dunno, 8th grade, is the following:
- if a player was a bit meh for many games and then he started to win more often then not, according to the score formula he will climb until a new level of score. The point is that the win ratio will not adjust itself quickly, why? well I leave this to the reader, but it is like answering to the question:
if I played 500 games and I won 40% of them, how many games should I play if from the 401th game I have a win ratio of 51% to show a win ratio of , say, 48%? Answer: a lot.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3390

Re: Grand master league?

Post#34 » 27 Mar 2017, 00:35

Another point.

http://gfx47.com/games/Gladiabots/Stats ... play=stats

You have at the moment 59% of winning score. Now either you improve quickly, and you get around 1750+, that means that you will be able to retain a good winning score, or your score will go under 50% pretty quickly, since you have only 132 games.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 565

Re: Grand master league?

Post#35 » 27 Mar 2017, 00:51

So to maintain a position you just have to spam match and win few time...

I know how elo works i play table tennis which use this system but in table tennis it's nearly impossible to stay at the same rank if you got less than 50% of win rate (not specially those with 49% but the guys who have only 40%)
Second point: table tennis use a system that reduce the inflation of points. Cause actually a player gain more points than the loser lose points. So at the end of each 1/2 years they calculed and reduce all the player by the general inflation.

pier4r
Skynet
Skynet
Posts: 3390

Re: Grand master league?

Post#36 » 27 Mar 2017, 01:14

TheKidPunisher wrote:So to maintain a position you just have to spam match and win few time...

I know how elo works i play table tennis which use this system but in table tennis it's nearly impossible to stay at the same rank if you got less than 50% of win rate (not specially those with 49% but the guys who have only 40%)
Second point: table tennis use a system that reduce the inflation of points. Cause actually a player gain more points than the loser lose points. So at the end of each 1/2 years they calculed and reduce all the player by the general inflation.



It happens also in gladiabots. Consider that you see mostly win rates that are cumulative.

So, let's solve the little problem.

You played 500 games with a win rate of 40% , then you finally improve and have a steady 55% of win rate. How many games should you play to show a 50% win rate?

so: 500*0.4 + x*0.55 = (500+x)*0.5
answer: you should play 1000 games more. So in total you will have 1500 games and 50% win ratio. Until then your 50%+ win ratio is hidden.

Second point, I am pretty sure that in your sport there are many players with similar ranks. Here too but mostly they are inactive, so it does happen often that people with very different ranks play against each other. If there would be more activity, having less than 50% of win would be not enough because it would mean that one would lose often with people with a score similar to his, therefore they would take a lot of points from him quickly and he will be demoted. This is not the case if only people with 200+ points plays often, this means that one would lose little points against them.

Score inflation is not a big problem really, because the score formula by itself cap the winning point after a while.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 565

Re: Grand master league?

Post#37 » 27 Mar 2017, 01:41

Oh ok. So some people lost a lot at begining and even if their %win is high for theirs last matches they still have a bad %win rate general. And people like me have to do more match to see their true %win.

It's more clear now thanks

User avatar
Ritter Runkel
Neural Network
Neural Network
Posts: 498

Re: Grand master league?

Post#38 » 27 Mar 2017, 21:16

Actually my winning ratio is somewhere between 80% to 83%. But with my score history the average overall is only 70%. So elo score gives much more information about the actual strength. Especially with more then 2000 points you often win and get no points because score difference is to high (but you lose allot when you lose a match).

Imho inflation is a point. Now we have more then six 2k+ players. I'm pretty sure it has to with score inflation.

pier4r
Skynet
Skynet
Posts: 3390

Re: Grand master league?

Post#39 » 27 Mar 2017, 22:04

Ritter Runkel wrote:Actually my winning ratio is somewhere between 80% to 83%. But with my score history the average overall is only 70%. So elo score gives much more information about the actual strength. Especially with more then 2000 points you often win and get no points because score difference is to high (but you lose allot when you lose a match).

Imho inflation is a point. Now we have more then six 2k+ players. I'm pretty sure it has to with score inflation.


Well so far the stats shows that the average of points of the active players in top league is not changing much. Therefore the top league is not harvesting a lot of points from the lower leagues, otherwise the average would change quickly.

Surely there is harvesting from lower leagues, but not that fast. See viewtopic.php?f=7&t=493

Instead I think that within the top league the points are accumulated mostly on the top end. Indeed if you see the distribution of scores, until 1800 it is mostly gradually increasing, and then there are big jumps. So it is like few players can collect from all the others or collect from the middle farmers.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 565

Re: Grand master league?

Post#40 » 28 Mar 2017, 13:46

pier4r wrote:Another point.

http://gfx47.com/games/Gladiabots/Stats ... play=stats

You have at the moment 59% of winning score. Now either you improve quickly, and you get around 1750+, that means that you will be able to retain a good winning score, or your score will go under 50% pretty quickly, since you have only 132 games.


You was right. I have spam match to reach level 7 quickly and even if my %win fall under 50% i am still at my average elo(1400)

Return to “General Discussions”

Who is online

Users browsing this forum: No registered users and 1 guest