Hunt rank algorithm - first 20 matches overview


#1

Two days ago I decided to use alternative account for my ranked Gorgon play. It created interesting opportunity to see how the rank algorithm works from early stage of ranked play. I am bringing it here - it is interesting, some results are obviously not in-line with the explanations (or expectations).

The data is not accurate. I was not making notes for all the results. However all the matches were played during last 2 evenings, I remember very well how it went, especially a few of those moments were the outcome was highly surprising. I guess - if needed for analysis - the actual data could be checked on server logs.

Notation:

  • I am using numbers to represent division, Bronze Skilled = 1, … Br.Destroyer = 5, Silver Skilled = 6, S.Expert = 7 and DR for “determining rank” players
  • M for monster division, H1…H4 for hunters division
  • the last column are estimates for points awarded to monster (win) or lost (loss)

I started with logging to my old account where I had just 9 matches (determining rank phase, most of them lost). I played 10th match, won, got initial rank as Bronze Expert, from this moment it went as follows…

M H1 H2 H3 H4
2 DR DR DR ~2 +40…50
2 DR DR ~2 ~3 +40…50
3 ~3 ~3 ~2 ~2 +40…50
3 ~4 ~3 ~2 ~2 +100…110 (wow!)
4 ~3 ~3 ~4 ~4 +40…50
4 ~3 ~3 ~4 ~4 +40…50
4 ~3 ~3 ~4 ~4 +100…110 (wow!)
5 5 5 5 4 -80…85

and from this moment - sudden change…
4 ~2 ~3 ~3 ~4 +1
4 4 4 4 3 +1 ?
4 3 3 4 6 +1 ?
4 4 4 4 5 +1 ?
4 3 3 4 7 +2 ?
4 ~4 ~4 ~4 ~3 +5
4 ~4 ~4 ~5 ~5 -4
4 3 4 4 6 +2
4 ~3 ~4 ~4 ~3 +4

then the strangest case - two matches in a row against THE SAME team:
4 4 4 4 3 +2 ?!
4 4 4 4 3 +25 ?!
5 …

It looks like algorithm defines two phases. First couple of matches rank was jumping like crazy. Until the first loss. BTW, it shows how is this possible that some players, having just 10…15 matches, went up to the highest ranks (and… stopped playing hunt to keep this high rank, I guess :confused:).

A few other moments were very strange. Being Monster div 4 couple of matches against div ~4 team for just +1 ? especially against teams including div 6 and div 7 players ?

And the strangest case - last two matches yesterday: against exactly the same team, division more-less equal to monster (4). First match just +2, next one +25. Why just +2 ? why increased for second match if their rank had to drop after first match ? I can see no logical explanation whatever ranking algorithm is used. That should not be possible.

I definitely suggest to review rank calculating algorithm or its implementation.


#2

Ranked works off a Glicko system it’s known to be ineffective hence why it’s been looked at in the past.

The Hunt algorithm doesn’t take the match into account. All it does consider is:

  • Monster Rank

  • Average Hunter Rank

  • Previous Matches (W/L Ratio)

  • Match Outcome

If you win in a minute at S1 you will gain no more points than you would if you win after 40 minutes at S3. It doesn’t look at the match itself but rather the general skill level of all the players, the match’s outcome and your last matches.


##Official TRS Response


#3

Funny thing is that I faced a team with a gold 1 on their team. I beat them and he lost 2 points. I immediately got paired up with them again and beat them but he proceeded to tell me that he lost 468 points and was now in silver division. I guess the ranking system determined him to not be gold material.

I was a silver skilled because I got disconnected in my placement matches a couple of times and I still beat higher silvers very consistently but I never get a lot of points. I occasionally lose to destroyers and about 2/3rds of my games against golds. Despite this I slowly worked my way up to silver master and now everyone dodges me.


#4

Just a second ago I had the following match:

me, monster: Bronze Destroyer Rank
hunters: 1 x Bronze Destroyer + 3 Silver Expert

So the team was having 1 hunter at my rank and 3 hunters two ranks above me.
The match was very good, but eventually I lost and I got … -40 points !

So how does it correlate to this explanation:

I was weaker two ranks and losing this match “told system” anything ? Was it any suprise for the system that I lost a match against two ranks higher opponent ?

I have no doubt that ranking system requires major rework. And if it is hard to adopt any 1vs1 systems - I suggest to change it to something very simple. Match won +X points. Match lost -X points. Very far form ideal, but fair and clear for everyone.

After all who is satisfied right now with how it works ? I mean some senior forum members try to be informative and give explanations “why”, but I do not feel they are happy either.


#5

Oh my… this kind of dramatic change should never happen, no matter what kind of match was lost - I mean in relation to typical delta about ±5…10 points. … 468 ? this is crazy.


#6

As has been described over in the other thread, your actual rank kind of bounces around within a much larger spread of potential ranks, what you are shown is what the system can confidently up to that point say it believes you are on average (for want of a better term).

It isn’t so much that your second match was different, but that by playing and winning that second match you may have bumped off and disregarded an older match which means that the system is now on the basis of recent results much more confident of your being a higher rank and so shows you a points boost.

The thing is the next match you could lose after this, one it expected you to lose, and it’d dock you maybe 3 points… but the system in the back is saying your rank is that level + or - a much wider range of potential ranks. It’s merely plotting a line, a trend of sorts, amongst that range.

I think that’s why people get these kind of “Whoa, why did I gain/lose so many points?!” moments, because they’ve played enough that the system says to itself “Right, we’ve been cautious long enough, I’m now confident that they are a grade above what I’ve been telling them” even though the trend line for your progression may have been smoother than that.

I think.


#7

Your explanation can be indeed correct. But if so - this is very unfortunate decision to differentiate so much results, based on some mathematical formula. I mean complex formula which “pretends” to behave like human judge and kind of “makes decision” which match is important and which is not.

BTW, how to make a vote about Hunt ranking ? I mean some people make such posts (I guess senior enough). I would ask question:

Are you satisfied with existing ranking system?

  1. yes, keep it as it is
  2. no, change it to another rank system, like ELO
  3. no, change it to very simple model ±X points for each win/loss

If this is not bad idea could someone post such voting, please ?


#8

I think you don’t quite have a grasp on how these thing have to work. There is no magic formula for being certain of what rank you are, all it can do is presume to know you are between a certain range of ranks, the range being smaller or larger based on how confident it is of it’s calculations being accurate.

It’s a feature that it is variable, not a flaw.

I’ve come to be completely satisfied with the ranking system in terms of how it ranks people, I’m only disappointed that there aren’t enough people to provide for the correct types of matches, and that some have decided to stop playing in order to preserve their rank.


#9

I understand that you try to give an explanation and this is naturally positive. But do we really have to have the most uncommon and the most complex ranking algorithm ? So many multiplayer games or tournamens use pretty simple ELO systems with variable delta. Variable - but within certain limits. People get slightly more if they win against strong opponent. Get slightly less if win against weaker. That’s it. Simple model. And it works. Works well in so many places. Why cannot we have something like this ? … no need for explanations and frustration for many players.


#10

It doesn’t matter what system is used, because of the way that it has been presented people would still get frustrated and feel the need for explanation.

The system works, that’s the thing that people don’t ever grasp when coming with these kinds of complaints you have. As long as there are enough people to match you up against similar ranked players then the matches you play against them are good matches, much better than Quickplay on average. But people are focusing on how much they are increasing or decreasing in rank rather than the primary reason this mode was made…to put people of similar skill levels up against one another to create better and less frustrating matches.


#11

I understand your point. However I have different opinion. I never heard people complaining for typical ELO model, nor I ever seen it generating so extreme differences like we have in Evolve.


#12

Hopefully with the ThunderChild telemetry system, 2.0 can get more data collected and consider how the game was played out.

I’d like for it to consider hunter comps and monsters. If the team comp is bad, less points are lost because you had a low expectancy to win, but if you win more points are rewarded. If you have a good team comp, you lose more and win less. Although I’d like to see the point system balanced so you can’t lose more than 20.


#13

I thought that as well, but he invited me to his lobby and what was gold 10 minutes ago at that time was now a silver. If the system thought I had no chance of winning and punished him hard for losing then I think I should have gained quite a bit of points. I think I only got like 10 points for that game.


#14

i’ve never been too invested in this game’s ranked mode as i think it’s a poor indicator of skill, maybe for the same reason you think it needs reworking. it certainly polarizes the playerbase. some people think it works just fine. i dunno.

:frowning: :frowning:

nice work on the self-telemetry though!


#15

So what if they implement thunderchild to gather a player’s info and rank them properly?


#16

While I agree the system works as intended, I disagree that it works within this game.

If the player base was still large enough to accurately match players with the appropriately skilled teammates or opponents, it would be great; however, it is near impossible, at least on Xbox One, to be matched with a team of hunters all within my league/skill level.

And before anyone says that, “that’s why you need to queue with a full premade,” it needs to be understood that that is a terrible argument.

The ranking system within the game is meant to queue me with similarly skilled players, and if it’s unable to do that, then it isn’t working properly in the scope of the game mode, whether or not it is working as intended. That’s the point I feel people fail to understand when they say it works.

I truly believe the ranking system used in Starcraft is the best system I’ve ever seen, comparable to the old Halo 2 in terms of accuracy in player skill ranking.

I have never gotten a game where I feel my opponents, or teammates, were not at a somewhat similar skill level, even recently during the period of time when the game’s population was relatively low right before Legacy of the Void was released.

I don’t know the math behind it, but it works around the active player population, dividing it into five portions, based off of their performance(somehow or another), then the top 8%, within the top 20%, allowing for hyper competitive games within the games own match matching service.

It’s also done in a season format, wherein the ranks are reset with each season, so it is malleable enough to allow for players to leave and join the population without disrupting the league divisions.

The first time you rank into a league there are 5 placement matches, similar to Evolve, but then with each new season, if you were already ranked previously, only one game is required to rank into a league.

A season format alone, even within the current ranking system used for Evolve, could do wonders as with each new season, players can more accurately be placed into their appropriate leagues, rather than being stuck within what other games refer to as ELO hell.


#17

I think this post will probably help you. Fair warning, it’s a long one.


#18

Yes, I have read it. But if you look at my telemetry and examples I gathered you will find many exceptions which do not follow the explanations given. For example:

  • two consequitive wins against the same team (I mean literally the SAME players) of equal rank to mine: first +2 (why so low?), then miraculously +25 (why suddenly more?)
  • loss against the team two divisions stronger than my monster which brought -40 ! points (why so much for losing against much stronger opponents?)