EPTSimple Questions, Simple Answers - Page 1117
| Forum Index > Dota 2 Strategy |
|
SilverSkyLark
Philippines8437 Posts
| ||
|
evilfatsh1t
Australia8930 Posts
On June 28 2015 13:50 SilverSkyLark wrote: Do you guys think that it's better if Valve used an ELO-like ranking system where players would get points regardless of the results? A win would give the max possible points and a lose would still give points but not as much. That's what KEPsA uses to rank Broodwar pros, as well as Magic: the Gathering (Planeswalker points are now based on an ELO System), and Chess Ranking. I think that it will make people less toxic in ranked games since they're still being rewarded. Unlike the current situation where people can really get toxic in ranked games because they got triggered. i dont know well enough about this system to comment but how do people lose points then? i seem to recall back when we had elo for bw pros the points did drop at the top and some ppl stagnated | ||
|
SilverSkyLark
Philippines8437 Posts
On June 28 2015 14:29 evilfatsh1t wrote: i dont know well enough about this system to comment but how do people lose points then? i seem to recall back when we had elo for bw pros the points did drop at the top and some ppl stagnated Good point. Actually I don't think people can lose points in that system. Maybe if they get into a losing streak they start losing points? I don't think Valve can effectively gauge a players performance based on his role to use that as basis for how much points someone gains/loses. | ||
|
Belisarius
Australia6233 Posts
Chess Elo, which you specifically listed, is zero-sum. The winning player takes points directly off the losing player. The loser loses what the winner gains, just like dota's +/- 25. My understanding was that SCBW's rankings worked the same way. As to whether using an inflationary system would help, I doubt it. For one thing, there are as many ragers in unranked as there are in ranked, and there are no visible ratings at all there. | ||
|
Birdie
New Zealand4438 Posts
And Elo definitely allows you to lose points. | ||
|
evilfatsh1t
Australia8930 Posts
it feel the tiers and divisions there are much more representative of skill than just numbers like dota. the smurf detection seems to be better there too. | ||
TheYango
United States47024 Posts
The advantage of the system is that it shuts up whiners who think MMR changes that are smaller than natural variation are actually meaningful, but for someone who actually understands the system to begin with, it's providing you strictly less information. Essentially, the system's sole purpose is to be opaque in a way that pacifies ignorant players who don't understand how Elo systems function. | ||
|
evilfatsh1t
Australia8930 Posts
the reason i thought lol's ranking system was more representative of skill was because in dota if a pro got calibrated at 3k for example, he would literally have to grind games to slowly climb the ladder. so you have a scenario where for the majority of this player's games, his rank is not representative of his skill in lol ive seen someone just last week, got calibrated into bronze 1 but played some games and next minute he was promoted to plat without going through silver or gold or even playing the promotion games (the player was a gold player in korea before starting a new account in aus, so he was pretty rightly placed). it seems lol's system of determining skill is more reliable/accurate than dota's | ||
|
SilverSkyLark
Philippines8437 Posts
I got a question about calibration though, is it based on your KDA after the 10 calibration games? Cuz I have a feeling that it disregards the role you played. | ||
|
Birdie
New Zealand4438 Posts
You play your first ten games on an account. The hidden MMR you are given will start at, say, 2250 MMR. The variance will be very high; that is, the amount your MMR can change after one game will be very high. Then, using some calibration algorithm involving damage dealt, KDA, last hits etc. the algorithm updates your hidden MMR. If you are smurfing, you will do very well at ~2250MMR and will have high KDA etc., making your hidden MMR rise very quickly (maybe even 250-500 MMR in a single game). After a certain number of games, the variance is lowered and you can only gain or lose the typical 25 MMR. All of this is hidden MMR at this stage. This is why you can jump into Very High Skill bracket games (check dotabuff) on a new account very quickly. When you start your ranked calibration, a copy of your current hidden MMR is made and used to determine your ranked MMR. I am not sure whether or not Valve modifies the variance during this period or not; it is possible that the ranked calibration matches are meaningless. Because of all this, a pro will never get calibrated at 3k unless he is deliberately throwing games. Once calibration is complete, the MMR gained/lost is based on the difference in team MMR. At most levels of play, this results in +/- 24 MMR, but at the highest levels, the team MMR disparity can be much higher because of lower population; when the MMR disparity is high enough, sometimes there is as much as +/- 45 MMR change per game. League of Legends uses a similar system. The leagues and promotions are just a front to make players feel better; most likely, they still match players based on MMR (which will typically be players within your league, but specifically players of similar MMR). | ||
|
TheYango
United States47024 Posts
On June 28 2015 20:00 evilfatsh1t wrote: how does the system work with promotions and calibration though the reason i thought lol's ranking system was more representative of skill was because in dota if a pro got calibrated at 3k for example, he would literally have to grind games to slowly climb the ladder. so you have a scenario where for the majority of this player's games, his rank is not representative of his skill Promotions are irrelevant because if your MMR climbs outside the MMR range actually represented by your division/league, you will get auto-promoted immediately. 99% of players never actually see this effect because their promotion rate keeps them within range of their real MMR, so they still stay within the MMR bounds of whatever division they're in the whole time. This was proven by people who would essentially deliberately lose their promotion series but maintain a positive winrate overall. Eventually, when they were consistently getting matchmade into games with higher division players, they got auto-promoted. So fundamentally, each division is still simply just a loose MMR range, and saying "I'm Gold 3" (representing, say, 1500-1700 MMR) is strictly less information than saying "I'm 1556 MMR". On June 28 2015 20:00 evilfatsh1t wrote: in lol ive seen someone just last week, got calibrated into bronze 1 but played some games and next minute he was promoted to plat without going through silver or gold or even playing the promotion games (the player was a gold player in korea before starting a new account in aus, so he was pretty rightly placed). it seems lol's system of determining skill is more reliable/accurate than dota's There are a two factors that have to be broken down here. One of them is an advantages of LoL's matchmaking, but is irrelevant to the actual league/division system itself: 1) In League, the initial division placement for a new account undershoots their MMR at the end of the calibration process. This is deliberate, but IIRC is actually not necessary anymore. Basically at the inception of the division system, the designers made it impossible to ever demote/drop down from a league due to poor performance. This resulted in a lot of people languishing at 0 LP at the bottom of a league (and encouraged account boosting because if a better player boosts you, even if you lose a shitton of games you still keep the league he boosted you to). The initial way to "prevent" this was to way undershoot new players' league placement and only bump them up when their MMR stabilized. So in your example, your gold player that got calibrated to Bronze 1 actually had an MMR in the Gold range, but had his initial league placement undershot slightly. As soon as he played a bunch of games and his MMR started to stabilize, the system determined that his MMR wasn't a fluke, so it promoted him accordingly. As of now, you CAN demote, so this mechanism is actually kind of a relic I believe. 2) The accelerated MMR increase/decrease of a new ranked account does not end after the calibration games. In DotA, your first 10 ranked games give you far larger MMR increases/decreases than prior games, and then you drop back to normal MMR gains which forces you to "grind". In League, you are placed into a League after the first 10 games, but your rate of MMR gain gradually drops down, and only if your winrate starts to stabilize--if you maintain a high winrate, your MMR gained per game will drop down to baseline more slowly than if you have a 50% winrate. Note that this effect actually probably *does* exist for a new account in its initial unranked games in DotA. The difference between League and DotA is that DotA uses unranked MMR as the starting point for an account playing ranked for the first time, whereas League begins from baseline MMR. In your "pro got calibrated at 3k" example, the reason is most likely not that he did poorly in his calibration games on that account, but because the unranked MMR of the account was relatively close to 3k. Because of this, League *expects* to need longer to reach an MMR representative of the player's skill, while in DotA, it's expected that an account's unranked MMR by level 13 is already mostly representative of their skill and the 10 calibration games are already something of a "second chance". On June 28 2015 21:12 Birdie wrote: Because of all this, a pro will never get calibrated at 3k unless he is deliberately throwing games. Well the more likely case is that he got calibrated to 3k because he didn't play the unranked games on the account and expected to be able to calibrate properly anyway. When you play your first ranked game, your MMR is set at the unranked MMR of the account, so if that's low enough, you can't possibly calibrate higher. The system isn't made to assume you gave the account away after reaching level 13, lol. Incidentally, smurf detection for new accounts is actually *better* in DotA than in League. It's just that smurf detection is not applied to ranked in DotA because DotA's ranked MMR uses the unranked MMR as a baseline and, again, does not assume that the account changed owners after the unranked games (or ever, really). | ||
|
njt7
Sweden769 Posts
On June 28 2015 11:21 SpiritoftheTunA wrote: when i see fata and synd zeus they often start just 1 or 2 clarities so that the bottle is pretty much guaranteed, then just bottle crow Lets say you are playing support Zeus and you get 1min of safe lane farm while your carry clears a stack. Would emptying your mana pool for cs be a viable option for you in that scenario? Trying to dissaprove a simple observation with specific scenarios from specific games of specific famous players isn't really productive imo. Sure you can learn from others way of playing but just mimicking without thought will hurt you in the long run. | ||
|
SpiritoftheTunA
United States20903 Posts
nor was i advocating mimicking without thought per se | ||
|
njt7
Sweden769 Posts
| ||
|
SilverSkyLark
Philippines8437 Posts
On June 28 2015 21:12 Birdie wrote: While unranked may use a slightly different algorithm (unlikely), the basic process is as follows: You play your first ten games on an account. The hidden MMR you are given will start at, say, 2250 MMR. The variance will be very high; that is, the amount your MMR can change after one game will be very high. Then, using some calibration algorithm involving damage dealt, KDA, last hits etc. the algorithm updates your hidden MMR. If you are smurfing, you will do very well at ~2250MMR and will have high KDA etc., making your hidden MMR rise very quickly (maybe even 250-500 MMR in a single game). After a certain number of games, the variance is lowered and you can only gain or lose the typical 25 MMR. All of this is hidden MMR at this stage. This is why you can jump into Very High Skill bracket games (check dotabuff) on a new account very quickly. When you start your ranked calibration, a copy of your current hidden MMR is made and used to determine your ranked MMR. I am not sure whether or not Valve modifies the variance during this period or not; it is possible that the ranked calibration matches are meaningless. Because of all this, a pro will never get calibrated at 3k unless he is deliberately throwing games. Once calibration is complete, the MMR gained/lost is based on the difference in team MMR. At most levels of play, this results in +/- 24 MMR, but at the highest levels, the team MMR disparity can be much higher because of lower population; when the MMR disparity is high enough, sometimes there is as much as +/- 45 MMR change per game. League of Legends uses a similar system. The leagues and promotions are just a front to make players feel better; most likely, they still match players based on MMR (which will typically be players within your league, but specifically players of similar MMR). Thank you for the thorough explanation. No wonder some calibration games are a walk in a park and some games suddenly are very difficult. | ||
|
starslayer
United States696 Posts
| ||
|
rebdomine
6040 Posts
Although I believe the way it's been set up in Dota 2 is that Techies loses the shallow grave buff before the damage to himself is applied. | ||
|
SilverSkyLark
Philippines8437 Posts
| ||
|
synapse
China13814 Posts
On June 29 2015 17:47 SilverSkyLark wrote: It would really be funny if you can suicide and not die under shallow grave. Kind of like how Wraith King's Aghs Upgraded ultimate when it was in beta. You can shallow grave the "Wraith" and then when the extended life duration ended and they were alive because of shallow grave, they'd still be alive. iirc you can still shallow grave them and then phoenix aghs ult them to bring them back. not sure if that was patched too. | ||
|
GentleDrill
United Kingdom672 Posts
I tested some other stuff that occurred to me too. If a hero suicides either with Suicide Squad or with Bloodstone suicide with an allied Aghs Scepter Wraith King nearby, they do not spawn a wraith. If, however, they are already a wraith, they will activate said suicide ability mostly as normal (dealing damage, putting it on cooldown etc. but the Bloodstone heal will not proc) but the wraith will not despawn until the end of the duration and can continue to act as normal during that time (Bloodstone heal will proc when they finally actually die). Pretty weird. | ||
| ||
