Autobalance or Elo

Greator_SST · Post by **Greator_SST** » Mon Feb 12, 2007 3:54 pm

ELO may never work because I think the range, 0 - 25 needs to be much larger, say, 0 - 100. Here's why. Consider a team where only 15s join vs. a team s where only 18s join. The difference between a 15 and and 18 is enormous, btw. Look how many stackers can join before ELO balances the equation.
ELO Team 1 ELO Team 2
15 -- 18
30 -- 36
45 -- 54
60 -- 72
75 -- 90
90 -- 108

Notice, that team two is stacked with 18s, team one has nothing but 15s. So, as always happens during autobalanced games since R3, and we've all seen it, the stackers still stack. Here we have two teams of six players, roughly equivalent ELO, but team 2 will win. Let's continue and see what happens.

105 -- (lo and behold, team 2 stackers are finally prevented from joining)
120 -- 126
135 -- 144
150 -- 162
165 -- 180
180 -- 198

So, by the time this game gets to 12 players, the only thing ELO has done to prevent stacking is to have given team 1 an extra player. Team 1 has 12 (15)s, team 2 has 11 (18)s, a pretty significant difference, especially given the fact that games are routinely stacked at the beginning which equates to a huge strategic difference.

You could keep playing this out and the balance doesn't really get any better. That's why I feel that we need to make the scale from 0 - 100 instead of 0 - 25.

spideycw · Post by **spideycw** » Mon Feb 12, 2007 4:24 pm

Psy speaks the truth...it doesnt work @ all (I foolishly turned on AA assuming AB worked)

Flower · Post by **Flower** » Mon Feb 12, 2007 4:27 pm

Greator_SST wrote:QUOTE (Greator_SST @ Feb 12 2007, 03:54 PM) ELO may never work because I think the range, 0 - 25 needs to be much larger, say, 0 - 100. Here's why....

You refer to Autobalancing by Ranksum instead of AllegianceEloRating aye?

Regarding the balance by Ranksum I agree that it would be beneficial to use the true AllegEloRating for AB as there is just no reason to use the coarser grained ranks for this task.

The Elo system in itself has no 'limitation' of achievable ratings though.
The spread of the ratings is solely determined by the playerbase, the depth of the game as well as the parameters of the system. (like the desired winchance chance with a given ratingdifference)
Sometimes when people implement the system a so called 'rating floor' is implemented. ('Anyone rated lower than X is shown to be rated as X' as well as 'Anyone rated higher than Y is shown to be rated at Y')
Such an adjustment is not part of the normal Elo system and is inadvisable as one is better served to adjust the spread of the rankings.

P.S.:
There is a difference between ELO, Elo and Elo rating system /tongue.gif" style="vertical-align:middle" emoid=":P" border="0" alt="tongue.gif" /> ;-)

Flower · Post by **Flower** » Mon Feb 12, 2007 4:28 pm

Btw. why is everyone expecting AllegElo to work to satisfaction yet?

Post Scriptum:
If it were not so horribly off-topic I would answer Spidey's question below in detail /mrgreen.gif" style="vertical-align:middle" emoid=":D" border="0" alt="mrgreen.gif" />
(I am in RT, isn't my weirdness and rambling about theoretical things enough of a clue? :-)

spideycw · Post by **spideycw** » Mon Feb 12, 2007 4:29 pm

I'm trying to figure out if you are in @Ace or @RT? ^^

Post by **Raveen** » Mon Feb 12, 2007 4:59 pm

Flower wrote:QUOTE (Flower @ Feb 12 2007, 04:27 PM) Regarding the balance by Ranksum I agree that it would be beneficial to use the true AllegEloRating for AB as there is just no reason to use the coarser grained ranks for this task.

The reason I can see is newbies, who start with an ELO of 1500 (ideal median of the system) but have a rank modifier based on games played/time or something. So they have an ELO of 1500 but a Rank of (0). Balancing by rank gives a better distribution of these players.

Flower · Post by **Flower** » Mon Feb 12, 2007 5:06 pm

Raveen wrote:QUOTE (Raveen @ Feb 12 2007, 04:59 PM) The reason I can see is newbies, who start with an ELO of 1500 (ideal median of the system) but have a rank modifier based on games played/time or something. So they have an ELO of 1500 but a Rank of (0). Balancing by rank gives a better distribution of these players.

Indeed True. :-)
(And if it would not give rise to the "Game is perfectly Balanced (according to ranksum) but does not count" issue I would agree)

Paradigm2 · Post by **Paradigm2** » Mon Feb 12, 2007 6:53 pm

I've used autobalance 3 times as a commander and left on autoaccept.

All 3 times, my team eventually had an 80% stack and the game didn't count... something is broken.

Dogbones · Post by **Dogbones** » Mon Feb 12, 2007 8:13 pm

spideycw wrote:QUOTE (spideycw @ Feb 12 2007, 11:24 AM) Psy speaks the truth...it doesnt work @ all (I foolishly turned on AA assuming AB worked)

We picked teams and it was a horribly stacked game... GASP picking teams does not work /glare.gif" style="vertical-align:middle" emoid=":glare:" border="0" alt="glare.gif" />
(and ergo, therefor, hence forth, we should never ever pick teams again /wink.gif" style="vertical-align:middle" emoid=";)" border="0" alt="wink.gif" /> )

AutoBalance, balances rank totals per team, period, nothing more nothing less. It does not take into account the comm's command ability, nor that a 25 may have a bad day, or that two 8's (in the same cadet class) may work exceptionally well together, nor a rix rush against giga scouts, etc, etc, etc.

It is a tool that can be used for pickup games where the coms are somewhat new and players not as well known.

It can also be used to prevent a horrible stack on one side. It will not prevent this in small games where a small 'stack' makes a huge difference.

We (as in Tiger and other's) will have a look at why things are getting to .80 / .20 when balanced ranks would suggest otherwise. If (0)'s actually have a 1500 elo, that is a problem.

Why don't we use the actually elo (as opposed to the rank) when balancing. That requires additional info being sent from ASGS to the lobby and then passed on to the server while maitaining backward compatiblity for those that do no run ASGS or have access to elo (as is the case for lan games, etc). This has not been coded yet.

Contrary to all of the 'hype' Auto Balance has not destroyed Alleg for all time nor has it solved the Stack problem.

If AB prevents you from joining a team when it is 3 down AND the rank totals of each side are somewhat close, let us know the what server and about what time it happened. We'll pull the logs. GASP there might just be a bug /blink.gif" style="vertical-align:middle" emoid=":o" border="0" alt="blink.gif" />

jgbaxter · Post by **jgbaxter** » Mon Feb 12, 2007 8:29 pm

BTW, doesn't the 'stats count' toggle in settings have absolutely no effect?

I know it was the plan that with AB on you'd toggle the 'stats count' in order to have them count, otherwise without both those things stats wouldn't count. Of course I realize that's for later once the kinks are worked out.

I simply noticed people thinking it did something and afaik it doesn't. /wink.gif" style="vertical-align:middle" emoid=";)" border="0" alt="wink.gif" />