Personally, I think that judging a players skill based on KPM and DPM is not an accurate measure. A lot of flaws into that system resembles the flaws reach had in his first season. If a player hogs power weapons/power ups, does that determine whether he's good or not? Why isn't assist not taken into consideration when it comes to measurement of MMR?
The system looks at your entire play history, so it can calculate a mean value for your skill across time. Negative outliers from this mean would be what the system sees as having a very bad match. Let's say you have once bad match, and just do really poorly. Your MMR might take a big hit behind the scenes, but luckily CSR can't shift as much as MMR, so one bad game won't send you from Diamond to Silver. Assuming you play normally the next match, your MMR will update back to around your normal skill value, meaning any CSR lost from the bad match should be made back quickly. Even a few bad matches in a row wouldn't lead to a permanent dip in rank. It would only be permanent if you kept having "bad" matches, because after a certain point it would be indicative of a loss of skill rather than bad luck.
I've had bad games and games, so of course I know the feeling, but even CSR doesn't forgive or care. It's a machine/system, but even then, how would it know?
I'm not really seeing "boosting" here. People have bad games/days. It happens.
My Onyx teammate underperforming not once, but twice and not getting punished because he's in a to2 with a lower ranked player
I thought boosting was no longer a thingI see Onyx/Diamonds vs Onyx/Diamonds.
Can you clarify what the issue here is?
Bungie, realized that error and made their system strictly wins and losses. This is a TEAM game. A player that averages 6 kills, but 10+ assist a game deserves just as much credit as the player that gets 17 kills and 5 assists with 5-6 deaths. It's an insult to players that are more supportive.
By what measure are you calling it less accurate?
We ran both a win/loss only based system on the exact same data full of millions and millions of players and matches and it was MUCH less accurate at predicting who would win each match than the system that incorporates KPM and DPM along WITH wins and losses. The win/loss system was barely predicting over 50% whereas the new system pushes over 70% on matches where we are already trying to make 50/50.
So no, I disagree with that. The new system is much more accurate at telling us who will win TEAM games, so therefore it is much more accurate at determining which players are good at winning, and therefore the ratings it puts on them are much more accurate representations of their skill.
The data also disagrees with your statement about assists. We've analyzed them heavily and players who get a lot of assists do not win more than those that don't. If you aren't helping your team win, then we're not going to add it to the model. Player who have higher KPM win significantly more than those who don't. The system is not designed to avoid insulting supportive players, but instead to educate them that if they want their teams to win more, they need to do a lot more than just Assists.
This isn't Reach's system. Reach's system was made by folks trying to guess at how things should be weighted and it was very inaccurate at it. They are night and day in difference.
This system uses machine learning and Bayesian Inference to exactly balance how much every factor contributes to a player winning a match. But the most important aspect of the system is still winning, not KPM and DPM. The final rating is still boiled down to a single number the same way it has been in every Halo. The new system just incorporates new data into the same win-based system.
Have you read the paper
? If not, you won't have enough information to make an informed judgement here.