[HotS] BlizzCon announcements discussion. - Page 2
Forum Index > Heroes of the Storm |
Corvuuss
Austria354 Posts
| ||
karazax
United States3737 Posts
On November 04 2017 20:44 KobraKay wrote: Not sure if I like the stealth changes.....but the rest seem like nice ideas to improve the game. Let's see if in practice they work. I'm just sad about Aba....no ammo means the benefit of his lane pushing will be greatly diminished no? They said minions would be getting health and dmg boosts to balance the lack of ammo, I expect something similar for summoning heros like aba/azmo/zagara. Also the removal of the tower behind the wall means one less tower shot at once to tank when pushing. They said one of their goals is to have heroes push with minion waves for value rather than clear and rotate as fast as possible. | ||
FeyFey
Germany10114 Posts
I just noticed the protector can dance. So many games were I didnt do that D: Alexstraza must be so overpowered to actually work. Since it really looks like your teammates have to be semi competent to make her work. | ||
karazax
United States3737 Posts
| ||
karazax
United States3737 Posts
| ||
karazax
United States3737 Posts
Regarding the new MMR system and possible abuse: The way the system functions, we aren't telling it what makes good play. It's simply measuring live games to determine what has proven effective for that particular situation. So, to "game" the system, you have to win the game while executing in that situation in a way that players have already proven is how to play at a high proficiency level. Personally, I'd love to have games filled with folks playing like that. The assumption a lot of folks seemed to hop to is that there would be some singular stat that they could focus on to abuse things, which isn't true. Every stat we measure is used as part of the calculation for every situation. It's just a matter of weighting as some are more important than others for that situation. If you were to, say, focus solely on soaking XP with the assumption that its the most important stat, you'd likely be under-performing in other metrics that balances it out. If you aren't under-performing in other areas, but are soaking more XP than expected for that situation...you're actually playing better. Win/loss is still the most important factor in MMR calculations: You'll be getting MMR adjustments based on individual performance whether you win or lose. You still only gain MMR when you win and lose MMR when you lose, but if you're performing beyond expectations for your current MMR, you'll gain more MMR on a win and lose less on a loss. The net effect is that, over the course of multiple games, you'll more quickly arrive at the higher MMR that's indicative of your skill. Then the opposite is also true if you're under-performing your current MMR. Why the xp bonus for selecting random shows up so rarely: We don't want to encourage people to select random normally, so the experience bonus only kicks in when the average queue time is above a particular threshold. The bonus is to then encourage people to play as a random so the system can more easily make matches and bring the queue time down. Combine that with ongoing efforts to improve matchmaking which has led to lower queue times in general and you don't see the bonus pop up very much anymore. Chance of getting a new map from the pool: The system doesn't currently support weightings on maps. It's just a random selection. That's why, in the past, we temporarily reduced the map pool when a new map launches so it can show up more frequently. That generates its own salt, though, so this time we just added it to the normal rotation instead. How a player can place at 1k master after their ever ranked games: To follow up on this, the placement was an interesting combination of things. First, you had a lot more games in QM than UD, but there was enough data in both QM and UD for the system to consider both valid choices to seed from. In that case, it uses UD as the seed since its more indicative of the play environment for ranked than QM is. Interestingly, your QM MMR is significantly higher than your UD one. Your QM MMR is master tier. UD is diamond. Both are above the placement cap of Plat 3, though, so that ends up not mattering much. What did matter, though, is that your uncertainty in UD was still relatively high since you don't have a ton of games played there so, since that was used for the seed, your uncertainty coming into ranked was correspondingly high. Then you won the first 6 games against high-skill teams, which bumped your MMR back up significantly due to the high uncertainty. Going 2 and 2 for the last 4 still left you with master tier MMR at the end of placements. It's not as high as your QM MMR, but a bit higher than your UD one. There's nothing inherently wrong with placing into Master as long as you're a Master-tier player, so the question then becomes: are you actually a Master-tier player, as indicated by at least your QM performance, or is it a fluke of the system? As you continue to play ranked, it should become apparent which is the case. Either way, this case was really useful to dig into and there are a few places along the flow that we may adjust things to improve the outcome in the future. Thanks for bringing it up! Why they removed placement cap, and wants to show MMR value directly: You have to be careful with anything that is artificially imposing limitations on where someone is ranked. Because rank is the only visible indicator of skill and people react allergically to folks of different ranks being in the same game together, even if they're the same skill, the matchmaking system prefers to match people of similar rank together. As long as their rank is appropriate to their MMR, that's a non-issue, but if they diverge too heavily, it causes problems. That's part of why we got rid of the Diamond 3 cap. It worked fine at the start of a season when there were a lot of higher MMR folks at the cap, but as the season went on and the higher MMR players moved up to master, the system had a harder time making matches for people who were higher MMR, but joined later in the season so didn't have that large pool of other, similar players to be matched with. The fix for this is making MMR visible, then decoupling MMR from rank entirely (and yes, that's on my wish list to do). At that point, we have a lot more freedom with rank without affecting matchmaking quality. Would you like to allow people to queue with preferred roles selected: No, but I would like people to be able to queue up as a preferred role and use that as part of matchmaking. It wouldn't be hard enforced, but it would increase the likelihood that you'd have a mix of people willing to play the different roles on a matchmade draft team. There is also a separate interview with BlizzClaudio here. | ||
FeyFey
Germany10114 Posts
| ||
karazax
United States3737 Posts
| ||
karazax
United States3737 Posts
Any changes also planned for uncertainty? Uncertainty doesn't exist in the same form in the new system. The core concept of something that limits how much MMR can adjust on a per-game basis is there, but the factors that go into it are different. Looks like new system promotes maining a few heroes rather than flexibility, true or not, and how big is the impact if true? The most important thing is still winning the game. That hasn't changed. Sometimes, that means sticking to heroes you know really well. Other times, that means having flexibility in your hero roster and playing the heroes that best fit the situation. Might be a stupid question, but if it's only about winning, why do we have stats-based MMR? why the win isn't the only factor? Isn't the concept of stats-based MMR less accurate than the winrate itself to calculate your odd of winning? It's about speed more than accuracy. The current system, based entirely on wins vs loses, works, but because your skill is diluted in a team of 5, unless you are dramatically better (or worse) than the rest of the team, it can take a large number of games before the difference in skill translates to more wins (or loses) and moves your MMR to where it should be. By factoring in individual skill to the equation, it can arrive at the proper MMR much faster which means the matchmaker can make better matches overall. Is draft/banning taken into account? Choosing the right hero in draft may be more important than performing better than peers on a certain hero. Drafting matters in the same way as the above statement. Drafting is a skill and how well you draft affects your win rate. All in all, do you have some additional insight into what parameters are used? How zoning, peeling, bodyblocking, skillshot blocking is taken into account? How will Force Wall-Tassadar be able to compete with Archon-Tassadar? We have great insight into what parameters are used...but aren't going into that level of detail externally. Also do you plan to add more parameters in the mix? We'll add more parameters as it makes sense. There's some concern about how reliable is data on low played heroes — maybe like Chen, Rexxar, TLV, anything on that? (Though I can't imagine any answer here but "we think our data is reliable enough" :D) We have more than enough data on all existing heroes. For new heroes or major reworks, the performance-based adjustments will basically be disabled until enough data exists. Does this mean that players of these heroes will receive smaller adjustments or larger ones? I'm afraid that if you are effectively docked points for playing unpopular or new heroes, that will become a self-perpetuating cycle. Neither and both? The system wouldn't be applying any modifiers based on performance during the time while its gathering the information it needs. It basically works like the existing system in that time frame, so if the player is playing better than expected (based on the future data) during that period, they'd be getting less points. If they're playing worse than expected, they'd be getting more. Are we right to assume that new MMR calculation would be applied to all game modes? For QM, it looks like a step in the direction of per-hero MMR, do you think we may end up there, or at least do you think it may be viable to maintain? It will apply to all game modes that have MMR. Can you win a game yet still lose MMR points? IE: a player goes full tilt, intentionally tries to throw. But the other 4 players are just dominating and still win the game. No. You'll only gain points on a win and lose them on a loss. If you get full carried for a game, you won't lose points, but you also aren't going to gain very many either. I think people's main concerns are along the lines of "who determines which stats are important" and "why these stats and not these other ones". I assume you guys are running some kind of adaptive machine learning, but most people are not going to understand that so maybe some kind of brief summary into how that works can help to alleviate the main concerns. Just a suggestion. Quite right. For the "who determines which stats are important", its the players. We've chosen the stats to monitor, but which ones are important for that situation is something we are measuring, not determining ourselves. About revealing MMR and the general clarity. Assuming "no" on "if you're going to reveal MMR" (:D) I'd like to ask whether we would be able to see when the adjustment takes place on the post-match screen, like we see PRA now. We won't have visible MMR for this update, but I do still want to do it in the future. For now, you'll mainly see the effects through rank points. We're planning to swap out personal rank adjustment for a more direct performance adjustment which will give better insight while filling a similar role of keeping rank close to MMR. I'm not sure right now whether that will be for this season or the next yet. If you show a player a number, he/she wants to maximise it or minimise it. While self-improvement (which would be visible by rising MMR) is something humans crave naturally (or so I've heard beep boop), visible MMR can lead to self-inflicted frustration and anxiety, like worrying over every MMR point lost or gained during a match. By definition, MMR is relative and obeys an ordering, i.e. I can immediately tell whether my MMR is higher than that of another player or vice versa. This can lead to acts of condescension towards players with lower MMR in one's games. Considering these points, I have always understood the leagues and divisions in ranked to be an abstraction of MMR to avoid these (and other) problems. What then makes you want to add visible MMR and how would you try to mitigate some of the more negative consequences of such a change? Spot on and part of why visible MMR hasn't been a big priority for us but, there are some advantages to it:
But what order of magnitude are you all thinking? 10 points? 20, 50? Just trying to get a rough understanding of the overall impact it might have on gaining. That's a tuning point we haven't settled on yet. The system, unless specifically programmed to do otherwise, will think a death at the last moment is extremely meaningful since it is normally a 60s death timer. Time is factored into the measurements. If you're dead for the last few seconds of a match, there wouldn't be any measurable impact relative to everything else that happened that game. Regarding the new performance based MMR changes, does this mean if my team has captured the winning altar on Towers of Doom and for the fun of it I walk into the enemy death zone and die, my MMR will be adversely affected from this little bit of fun? Or perhaps there are 5 enemies killing my core and there are 4 allies dead. I have no chance to defend the core but rather than try and risk receiving an extra death I won't lose as much MMR if I stay in base? Curious about this. Everything is by time, including deaths. If you die in the last second of a match, its not going to mean much. Even a full death in the last minutes would really only be a minor modifier overall once diluted against your overall performance in the game. The biggest factor for MMR adjustment is still whether you won or lost. Your overall performance is secondary to that and any particular moment is just a small part of that overall performance. Does this mean that winning a game quickly and playing well will reward the same personal adjustment as having good stats in a long game? Some people are concerned that players will want to artificially extend game lengths to try and pad their stats before winning. Especially in games where the opponent is greatly outmatched. Yeah, definitely. Everything factors game time into it for that exact reason. Anecdotally, they tested this on a Master-level player - but how does it work for below-average HL players (bronze and silver)? It's been running on live servers in parallel with the existing system for the last couple of months. We've been looking at tons of data. The Master-level example I gave at BlizzCon was just one I thought would help most folks grasp the impact of the changes. One thing I'm interested in is whether the system is trying to distinguish between stats during a loss and stats during a win. It probably is, but as a support main, it's important. It's really hard to get high healing numbers in a losing game, because losing games often involve people not wanting to live very long. You can't heal very much when your Arthas re-engages at 20% health and that's the end of the fight. Yes, the comparisons for winning games are different than for losing games. (side note: I'm curious how the system evaluates a "good" performance in a lost game if it's not comparing the performance to the stats of winning players?) Khaldor is going to do a Q&A video interview with BlizzTravis with more insight so you can post any questions you want to ask about it there. | ||
karazax
United States3737 Posts
| ||
| ||