Community > Community Events

Cronus League - Single Ship Teams

<< < (2/5) > >>

nanoduckling:
Had a look through this, how useful my comments are depend on what you are going for with the tournament. First thing I'd note is that the point system is somewhat similar to Elo (the rating system, not the band), more complex in some ways and simpler in others and achieving slightly different goals.

The reason rating systems aren't used for tournaments is that they weight recent success or failure more highly than distant success or failure to estimate how good a team is. This makes them different from a standard league where early victories count just as much as later ones. They aren't as extreme as a knockout tournament though where early failure gets you, well, knocked out. The virtue of this point system being like a rating system is that it implements a hidden catch up mechanic, where a team with less points is risking less when it competes, and gets more rewards when it succeeds.

It isn't as extreme as Elo in that regard; typical ratings changes in Elo are on the order of about 16-32 per game on a score of 1500 and typical changes here are 100 on a score of 1000. Points totals will change faster. One thing Elo does that might be worth considering is bigger points changes in the early stages. We know that a points total of 1000 is very wrong for many of the teams likely to compete, so it might be worth making say weeks 1 and 2 have higher points changes, this could also be used to shorten the tournament a little.

I think the score system is good as is and probably wont benefit from substantial changes.

Team pairing is a different issue. Pairing teams this way could have some nasty consequences. Say team A is amazing, team Z sucks, team B is decent, and team C is decent but not as good as B, league is currently in order and these are the only teams. A+Z gets beaten by B+C and now B is top, A is next, C is next, Z is bottom. Next match B+Z is beaten by A+C and we swap right back round again. Who comes top is likely going to be determined by if there are an even or odd number of weeks in the tournament.

Problem is most of the methods I can come up with have scenarios comparable to above. Say we have completely random matches each week (not just opposition), well if the dice land poorly in the last week then we can end up with a team getting points by being paired with better teams at random. The best I can come up with is to vary who you are paired with within a week. So first match is say A+Z vs B+C, second is A+C vs B+Z in the example above. This has the disadvantage that practice is a bit of a pig since you will have multiple teams to work with, but it removes some loops in score like the above (not all of them though).

Another option is to just have a computer run every possible pairing and play those sets of matches with the smallest differences in total points scores (the idea being to make it so teams are risking roughly the same amount of points). There might be enough variability in there to avoid these kinds of issues.

In any case looking forward to seeing how this develops, good work all involved.

Thomas:
Some great comments, but could you clarify the 'We know that a points total of 1000 is very wrong for many teams likely to compete..'? I'm not sure what you mean there. The point value is pretty arbitrary, as is the % of points lost/gained for each match. Something we can change pretty easily. One of the reasons the tournament is kind of long is to prevent a single week or two from determining the total outcome of the league. In a shorter tournament, if you have a single bad week or so, you might never catch up.


I'm not as sure about making the early matches worth more, particularly the first week matches. This is mostly because we have the early sign up bonus of being allowed to pick your ally in week 1 for the first match or so. This will likely lead to some pretty imbalanced matches, which is one of the reasons we want some random pairs as well. Week 1 is going to be an odd week.

What's kind of neat about the points in this system is that we're pairing two teams together. So a team with a lot of points and a team with a lot less points. At the end of the day, each of the pairs will be roughly around the same amount of points, so all teams have the potential to earn about the same number of points. We're trying to reward teamwork. So we'll have a team with 1750 with a partner who's at 400 going up against a pair that has 1200 and 900. Right now the different pairs for the matches will be random, but we could look into setting them up in a way so that pairs don't always fight the same people over and over or so that they go against pairs with closer point values.

MightyKeb:
I think the idea of a different ally each week in a 10 week tournament sounds distressful, atleast to me personally, unless I've read it wrong. What I would suggest is somewhere halfway the tournament (week 5-6) each ship has their amount of points counted and placed into similiar tiers as those that exist in Blood and Brass. From here on, you can either randomly assign ships to other ships based on this tier or have them pick their allies themselves, or something between both. However, at the end of this, each ship stays with their paired ally from week 5-6 up until week 10. What this allows is for two teams to build relations, familiarize themselves with eachother's playstyles and build/practice on it to become even deadlier, thus elevate the skill level of a team itself and make for much more competitive matches.

nanoduckling:
Oh what I mean by 1000 points being 'wrong' is in the sense of if it is viewed as a rating system, which it sort of is. Give me nothing but the scores of the teams going into say week 8 and I can probably predict a decent number of the results of week 9. In week 1 the numbers are meaningless (since they are all 1000), so between week 8 and week 1 the matches add information to the ratings. The absolute value is completely arbitrary as you say (the starting value is just a number after all), but those numbers will mean something in the later weeks. So what I mean is if you want them to contain more information earlier then you can adjust them more aggressively early on since you know that initially the scores are a better reflection of team skills than the rating are. Later on this wont be the case so more conservative adjustments would be appropriate.

On the other hand it might be disheartening to have a massive chunk of points taken in the first couple of weeks so statistical purity should probably give way to competitor psychology, and the first week is a little odd and should almost certainly be an exception in any case as you say.

The virtues of the point system proposed are as you suggest, but I don't think I've made my concern with the pairings clear. It isn't just who you fight that is the problem, it is who your partner is. The bad scenarios I think it is worth avoiding are 1. A terrible team get paired with the same team every week, crippling them 2. A terrible team gets swapped back and forth between two other teams, whoever is holding the hot potato in the last week loses 3. Chance plays a major role in the last week where a team gets an objectively tough draw due to randomization.

Thomas:
The reason we change partners every week is so that we keep things fresh. We could just put teams together and keep them that way, but that feels like it ruins the point of having a 4 person team. You might as well have a regular tournament with a pairing system tossed in. One of the fun things of new teams is also being able to find out which 4 person teams play well with others. It's not just about being the best team, to get there you have to be able to work well with a variety of other teams.



One thing we can do is make week 1 just for ranking purposes and have no points exchanged.

Yeah, partners are a big concern. With the teams earning points together or losing points together, they tend to move up or down the board together, and makes it less likely for them to be paired repeatedly. It's not impossible, and if it becomes a big concern we can add a clause to prevent this from happening, as well as preventing/reducing the flip flopping.

We do have the top 4 teams moving into the final rounds, which should mitigate some of the random factor in the final weeks. The team with the most points might not necessarily be the best, so we get the top group to find out who's actually the top dog. We could up the final group to 6 or 8 to mitigate it even more. It would take a little longer, but it might actually work out better, with less chance of ties.

 

Navigation

[0] Message Index

[#] Next page

[*] Previous page