Fair points. Any chance you could point us to some of those academic papers, or other recommended reading in this field? I'm genuinely interested by this topic and would love to read more about recent innovations and research. I've got a masters in applied math (ahem... nerd), but my job is much more focused on statistics than anything specific to ranking systems.ZaedynFel wrote:If you mess with Elo’s math to force Onyx to be smaller it’ll mess everything upI don’t need to touch the math, just the boundaries on the raw numbers that are used to decide your ranks. For example, right now, 1500 is the boundary for Onyx. Maybe that needs to be 1700. But your raw CSR and the math behind it would stay the same.
As a side note, I reserve the right to mess with Elo’s math as much as I want since I've published academic papers in that type of math(s).
Thanks. I really appreciate that you're looking into doing something to discourage smurfing (which has become a really annoying problem in multiple playlists - I match smurfs once every ~5 games) - your suggestion of forcing CSRs towards the fireteam average after every match is reasonable, since halo is SUCH a team-oriented game that it seems near impossible to accurately determine the relative contributions of two players in a game when they have dramatically different skill levels. So, slowly forcing both players' CSRs towards the team average emphasizes the team aspect of Halo. I think the "downside" you mention is tolerable if it decreases smurfing... If you're legit playing with worse players but not boosting, you probably don't care much about your rank anyway! Plus you're playing against lesser skilled opponents so it's not a fair assessment of your supposed champ-level skill to run train on plats. ;-)ZaedynFel wrote:Not Using the Average Skill in PartyI’ve heard people suggest that we should use the max skill of a party instead of its average in order to discourage partying up with smurfs. I’m not opposed to that and may try it in the future, but:
- I want to first to see how well the skill improvements alone can mitigate this problem.
- If still needed I may do it for only Team Arena
- A softer option is to force your CSR to move towards your fireteam’s average after every match rather than your personal MMR. For example, in the case of an Onyx player playing with Bronze, both will see their CSR drift towards Gold over time. This will have the result of discouraging smurfing since it will drag the better player’s CSR down. In cases where your friends really are bad, it also will still give the more balanced matches for everyone that we have today. It does have the downside that when you play with bad friends, your CSR goes down. But so does matchmaking on the max. I think this option is better overall.
So are you saying that our current MMR/CSR ranks are based both on wins and on individual performance stats, but a future idea is to separate into two CSR numbers, one based solely on wins (Elo-like), the other based solely on performance stats? I would strongly support that.ZaedynFel wrote:Team Rankings vs. PersonalYes, I like the idea of having a Personal Ranking that moves based on individual in-game performance, and one that is based on winning. I would want these both separate and want to make it clear that the winning version is more important.
Keep in mind we don’t use Arpad Elo’s method for Ranks. We use something more advanced. We can integrate Kills, Deaths, Assists, Headshots, Objectives, etc., into your skills. We can also create a win-based one CSR separate from your individual performance-based one. I’m constantly evaluating which would be best for each part of the game.
I'm a huge skeptic about incorporating individual performance stats into primary MMR skill ranking (and therefore influencing CSR) in a team game like Halo... I know you currently do it that way, and this is an age-old debate. But why do you feel so strongly that incorporating performance stats is necessary, when Halo skill is SO situational and performance stats never tell the whole story? I've watched my games in theater mode thinking "I went HAM that game, how the heck did we lose?", and discover that my higher-ranked teammate, who had great stats, cost us the game by playing selfishly/conservatively. In one case, he even had awesome objective stats (lots of stronghold defense/secured - although turns out none of them were at critical moments), but in watching his gameplay I discovered that he kept letting us die and preserving his own life by not getting in the strongholds to stall or maintain progress at critical moments-- we lost 100-95, largely because of him.
Often, the "best" player on the team is taking on more risk, aggressively challenging the key map positions / weapons / powerups, and may have worse stats! I see this in practice all the time - I'm a Diamond level player but often end up with better post-game stats than onyx teammates when solo-queuing, because they're challenging for map control and trying to make my life easier.
I wrote Python software to scrape the Halo 5 API game histories and player ranks, to generate these bell curves and "percentiles" to go along with them. It took a lot of time to scrape the API, so I stopped doing it after a few months last winter, but the response on Reddit was really positive! I think a lot of folks would love to get a sense what percentile they fall into, so I'd love to see official data on this matter if you are able to publish it at some point. For context:ZaedynFel wrote:Can I show the bell population of the ranks?I won't at the moment, but since enabling demotions, it’s smoothed out quite a bit. The amount of players within each rank among those who have completed placement is at the intended distribution, except for perhaps Onyx being larger than intended, which I spoke on last week.
Thanks for taking the time to post so many of your thoughts for community discussion. Any specifics on when the changes will go live?
