Support NeoGAF

[Brawndo Addict]

The problem with the Hotslog talent data is that it really only tells you what's popular (or due for a nerf). Nearly every talent is going to have a winrate between 45-55% to begin with, and once you factor in a margin of error it's basically a 50/50 wash. Sure, it can help you identify 'trick' talents which you're not supposed to take, or 'OP' talents they haven't tuned yet, but those are few and far between and you could probably figure those out already. Their talent build section is incredibly, incredibly misleading because the win-rate percentages reflected there are not accurate whatsoever. Obviously a hero with an overall win-rate of ~50% and a individual talent win rate of ~50% cannot have 10 talent builds all above 60%, it just doesn't add up. The reason this happens is subtle but obvious once you think about it.

The reason why is because it's restricted to talent builds that go to level 20. There are more games where the winning team hits 20 and the losing team does not than there are games where both teams hit 20. This is why you can see a winrate spike towards 60% for the individual level 20 talent tier for every hero, it's reflective of those teams having already won their games, not because of the talents themselves. Even a 'bad' level 20 talent is a gigantic power increase over an opposing team without level 20. This effect filters down into the talent build section because they only include builds that reach level 20. So all the win-rates for those builds may say they're above 60%, but it's because they're reflective of the data being overpopulated with lopsided games.

I think it's more important/useful to use Hotslog data to think about why certain talents are so popular and how they interact. Take Xul's level 4 talent tier for example. When you look at the individual talents, Jailor (which spawns 2 Skeleton Warriors with Bone Prison) is ahead of the other two, increased range on spectral scythe and gain mana when trait activates (although they're both close to 50% too I might add). Why is that? Well if you look at the talent build section, you can see that it's very popular to take Jailor in conjunction with the trait enhancing talents at 7/13/16. Since Jailor directly synergizes with those talents, it gains more value as the game goes on. So by level 16, Jailor doesn't just summon two extra skeletons. It summons two extra skeletons who heal you, explode, and do 50% more damage than they normally would.

Many heroes have talent interactions like this involving similar feedback loops. But it's also important to remember to take into account the map you are playing on and the enemy team you are facing. Giant Killer talents (extra dmg based on HP pool) gain more value if the enemy team has more warriors or high health heroes. Extra range on ability talents are more useful on maps where you can poke over static objectives. And some heroes are just plain better/worse at certain maps (all of Xul's talent win-rates go down on Battlefieds of Eternity for example). By thinking about how a talent changes the way you play (either based on the map, enemy heroes, or just your own kit), you can make better decisions over whether it's the right choice for that game. Just because something is the "best" build overall doesn't mean it's the best build for that map, for your composition, or against the enemy team.

 

[brian!]

I find Kael to be bad tbh. You can stack his flame strike quest really fast (I had it at leve 7 I think), but that means no arcane barrier and you still have the problem that people just move out the flame strike, and even when they don't it's not as much damage as a landed blizzard or li-Ming orb.

Have to be honest, I'm more dissapointed with the new Kael everytime I play him. I just don't get how they felt this was in any way balanced or even fun compared to everyone else out there.

for me heroes like kael or nazeebo have a really hard time dying because they just chill behind everything and never need to put themselves in danger, he's def one of the best in my eyes for aram esp on this style of map w/ tight areas and bad flanking

 

[Maledict]

My issue is he just doesn't get much done compared to Li/Ming, Jaina, Falstad or Valla. Especially li-Ming, she's atrociously powerful it feels on that map. Boom boom boom!

 

[jaundicejuice]

ARAM is the only place that Kael's level 20 Living Bomb talent is probably worthwhile.

 

[Fixed2BeBroken]

Worth noting that if you pick up the Origins edition from Best Buy, you get free $10 credit.

It's kinda like getting free Tracer.

Kinda.

you sure about the $10.00 certificate? i dont see anything about that?

 

[Brawndo Addict]

Would convection stacking higher make it a more viable option? Maybe make it so you keep the stack every time you hit +100 damage? So it resets to 0 on death until you hit 100, then it resets to 100 until you hit 200, etc.

 

[Alur]

Lots of stuff concerning hotdogs

I see all of this, but I feel like you're just reading way too into it man.

If you check out any tier list, just about any build during a competitive match, guides on the various websites, and hotdogs data, they almost always mirror each other 1-for-1. There's very little variance bar a tier or two on most heroes in most games where people are trying to min-max.

If you're in the market for a build and either can't or don't want to figure it out yourself, there's not really a better tool to be had that has any kind of numbers behind it. Most every hero has one specific build that everyone plays, bar a tier or two, and hotdogs reflects that exactly. If that's what you're looking for, that optimal build, then hotdogs is the best place to find it.

If you're someone like brian! or yourself, apparently, then maybe not. Theorycraft it up on your own. There's nothing skewed about the stats, though. They are what they are. Level 20 games, not level 20 games, there's more than enough data on virtually every hero and 98% of talents to come to a reliable conclusion about what talents are good and what ones are bad without having to think about it too much.

Someone who is just jumping on a hero the first time and willing to go find a guide or what have you probably isn't worried about the map specifics, or matchup specifics, is my point. Hell, 80-90% or more of the HOTS playerbase are not interested in that based on the data we have.

 

[Fixed2BeBroken]

I think it's more important/useful to use Hotslog data to think about why certain talents are so popular and how they interact. Take Xul's level 4 talent tier for example. When you look at the individual talents, Jailor (which spawns 2 Skeleton Warriors with Bone Prison) is ahead of the other two, increased range on spectral scythe and gain mana when trait activates (although they're both close to 50% too I might add). Why is that? Well if you look at the talent build section, you can see that it's very popular to take Jailor in conjunction with the trait enhancing talents at 7/13/16. Since Jailor directly synergizes with those talents, it gains more value as the game goes on. So by level 16, Jailor doesn't just summon two extra skeletons. It summons two extra skeletons who heal you, explode, and do 50% more damage than they normally would.

I just use hotslogs most popular talent builds as a reference...so for instance..

If I'm learning a new hero....I go with the cookie cutter hotslogs build just to get a feel for the hero....then as I play more games...I start tuning my build based on comp/map as I gradually get a understanding of that hero.

For instance with falstad, I started out with the most popular build based on hotslogs, but now my 1/10/13/16/20 all vary depending on comp/map. (that 7 never changes doe)

I'm assuming others do this right? no?

 

[Familienoberhauptvogel]

Hotslogs is raw data what you deduce out of it is your own. The tools are there to look at it more selectively.

I'd love for there to be an indication of certain picks against certain heroes. Say Li Ming picks force armor 70% against Jainas

 

[Alur]

I'm assuming others do this right? no?

I would imagine most regulars here do. I do not think that most people who play this game do by any stretch, though.

People go for the path of least resistance. Most people don't take this game as seriously as we do here, though, bantering back and forth about how many seconds added to Gust is too many to make it useless and shit like that lol

I'd love for there to be an indication of certain picks against certain heroes. Say Li Ming picks force armor 70% against Jainas

That'd be neat to see. I'm sure he's capable of doing that if he wanted to.

 

[kirblar]

you sure about the $10.00 certificate? i dont see anything about that?

It's on the list of $10 bonus preorder titles. If you have GCU, getting the physical edition instead of the digital is a no brainer.

 

[Fixed2BeBroken]

It's on the list of $10 bonus preorder titles. If you have GCU, getting the physical edition instead of the digital is a no brainer.

cool, didn't see it on the product page.

 

[brian!]

My issue is he just doesn't get much done compared to Li/Ming, Jaina, Falstad or Valla. Especially li-Ming, she's atrociously powerful it feels on that map. Boom boom boom!

yeah li ming is nazty

 

[Alur]

Is KT still good in ARAM without the old chain bomb? Also do the bonus orbs stack his questing talent?

 

[Brawndo Addict]

I can't disagree on the appropriateness of complex information when you're just looking for quick and casual instruction (and in that respect you can consider the originating discussion as a launching point for a broader subject in a niche community). But I do disagree about how inherently objective the information of things like Hotscanine is, or rather, how useful it is.

It's important to remember that it is not the raw data, it is a subset of selectively provided data. The raw data is at Blizzard, what we have is the data from a particular group of players who upload their games to a third party site; it is not randomly selected or sampled. It's why you had Blue Posts reminding people that their MMR on hotslog is not their actual MMR with respect to who is chosen as the Hero League Captain. It basically just means that it is vulnerable to sampling errors and this is rather pronounced when looking at level 20 talents and thus, win-rates of talent builds.

I'm not saying the data they have access to doesn't actually reflect the numbers they are putting out there. What I'm saying is that the way the data is naturally produced leads to these statistically misleading numbers. It's why that predict a winner based on team comp/talent builds thing had to hide the talents after level 10. If you knew that one team hit 20 and the other team didn't, you knew who the winner was. It's the same principle here.

I'm not sure how to balance that problem as I'm not a professional statistician, but if you're trying to evaluate the effectiveness of particular level 20 talents on whether it helps you win the game, you have to filter out the games where your level 20 talents didn't matter strategically. Because the talent builds section is exclusive to builds that reach level 20, they have the same disproportionate results. They're accurate in the sense that they were calculated correctly, but they're inaccurate in the sense that it's not providing the information you want to know because of how the data is generated. 65% of players do not win with the build listed. 65% of players whose games who have been uploaded and who reached level 20 won their games with that build.

 

[brian!]

yeah but id recommend the q talent it's more fun and you can stack it by like lvl 3-4 so yeh instant +150 in an aoe while azmo's like brooo...ur making me jealous

 

[Alur]

It's important to remember that it is not the raw data, it is a subset of selectively provided data.

People argue this all the time and I just don't buy it based on the huge amount of data we have had over months and months, coupled with the sources from Blizzard which have corroborated most of the data to a large degree. Is this an "people only upload their wins" kind of thing you're implying, or what? Cause there's 9 other people you have to rely on not uploading to game it in that way. Beyond that, with the kind of sample size we are dealing with it's about as reliable as data gets for this kind of thing without Blizzard giving us their own, which they will never do. It's corroborated in most cases by the in game draft/ban system, and has been corroborated multiple times by the leaderboard released by Blizzard.

it is not randomly selected or sampled.

Why does it need to be? It's a mass dump of what talents people pick the most, how often they win with them, what maps they win on, etc. People have pulled out selective dumps trying to confirm or deny long held beliefs like "the team with the lowest MMR player is more likely to lose" and all of that. They've found barely any deviation in those dumps compared to the whole pile of data.

but if you're trying to evaluate the effectiveness of particular level 20 talents on whether it helps you win the game, you have to filter out the games where your level 20 talents didn't matter strategically. Because the talent builds section is exclusive to builds that reach level 20, they have the same disproportionate results. They're accurate in the sense that they were calculated correctly, but they're inaccurate in the sense that it's not providing the information you think it is because of how the data is generated.

Ultimately, though, beyond the hotdogs may be/may not be a lie conspiracy theory that we've had going on since Alpha, this is the part of your posts that has flummoxed me.

I'm not, and I don't believe anyone else is talking about looking at the completed talent build section. When I suggested it to the poster, I was referring to the actual majority of the page (http://www.hotslogs.com/Sitewide/HeroDetails) where you can see each individual talent broken down by number of times picked, winrate when picked, and overall popularity for all brackets or for your given bracket or anything in between.

If you're trying to find a build, there's pretty much no more reliable way than to look at that and see that yes, not many people pick Totemic Projection at 7 on Rehgar. As Familie said, you can do with that information what you will, but odds are over a large enough sample size there's a reason certain talents are completely neglected.

They're accurate in the sense that they were calculated correctly, but they're inaccurate in the sense that it's not providing the information you want to know because of how the data is generated.

Also, what is inaccurate about how the data is generated again? Uploading only wins or something like that? The sample size is just too large for that man. You can go through every hero and look at the most picked talents, or hell, even two most picked talents on each tier and get near universal agreement on them being the correct choices in most scenarios. What exactly is inaccurate or misleading about that?

 

[Zackat]

I use this http://www.hotsbuilds.info/ neat site and easy to navigate. It updates from hotslogs pretty frequently and gives you other sources to get an idea of talent builds.

 

[Alur]

I use this http://www.hotsbuilds.info/ neat site and easy to navigate. It updates from hotslogs pretty frequently and gives you other sources to get an idea of talent builds.

Haha, I forgot about that site. It pulls my HeroesFire guide on Nazeebo for the HF build on it's Naz page. The first version I posted of that thing was in December of 2014.

 

[Zackat]

Haha, I forgot about that site. It pulls my HeroesFire guide on Nazeebo for the HF build on it's Naz page. The first version I posted of that thing was in December of 2014.

Old school nazeepo plays, and to think he used to be a meta-pick hero. He still good at pubstomp I guess.

It's pretty nice to have everything there at once with the guides and everything for newer players. I just use it to refresh my memory on heroes I haven't played for a while. I haven't seen anything better than it so still using it from when I saw it on reddit a long time ago.

 

[Alur]

There was another site that was supposed to show counters to specific heroes that were voted on by users in a reddit style upvote/downvote system. I have no clue what it was called, though, or what happened to it.

Feels like we've had 5-10 of these that were good ideas or useful to some degree but I never bookmarked most of em. I still remember when hero.gg first dropped and people actively uploaded to both hero.gg and hotdogs for a bit. It looked a lot better visually, but it was kinda buggy. iDream and that Starcraft dude were always listed as the top two most searched players...I forget the SC dude's name, but he left the scene quick.

 

[Brawndo Addict]

People argue this all the time and I just don't buy it based on the huge amount of data we have had over months and months, coupled with the sources from Blizzard which have corroborated most of the data to a large degree. Is this an "people only upload their wins" kind of thing you're implying, or what? Cause there's 9 other people you have to rely on not uploading to game it in that way. Beyond that, with the kind of sample size we are dealing with it's about as reliable as data gets for this kind of thing without Blizzard giving us their own, which they will never do. It's corroborated in most cases by the in game draft/ban system, and has been corroborated multiple times by the leaderboard released by Blizzard.

I'm not trying to argue a conspiracy. I'm just pointing out what should be uncontroversial statistical terminology. Data is selectively uploaded, by that I do not mean to imply ill will, merely that the site is dependent on players themselves providing the data. The average player does not do that. The average player does not even use hotslog or forums like this. This means that the data you receive is going to be skewed based on the population providing it to you.

This is balanced out a bit by the fact that one person can upload results for the other 9 people. But uploaders are not equally or randomly distributed across all skill tiers nor are the populations of players in those skill tiers equal. People who are more skilled or into video game culture are more likely to upload their results than others, so your results will be skewed to that demographic. And even if things were perfectly balanced, you would still need on average 10% of your players to upload results which is enormous.

Will the results be close to the full data available to Blizzard? It depends on what you mean by "close" and it depends on kind of data/information we're talking about. If you mean, within 5-10%, then sure, it probably is. But an margin of error of 5-10% makes the information pretty questionable in my opinion if you want to know something more than which talent is the most popular, that is, if you want to know something about win-rates and similarly derivative information.

All the useful information is going to be between the 40-60% range, probably with another cluster in the 45-55% range. Getting your margin of error down low enough to work in that range is really hard without random sampling or the full data set at your disposal. I don't doubt that Hotslog is in the ballpark for a lot of these figures like MMR. But being off by 10% for MMR is a lot different than being off by 10% for a talent's win rate. So yeah, when we're talking about winrates that are being intentionally tuned around 50%, the inherent margin of error problems become a lot more serious of an issue.

Why does it need to be? It's a mass dump of what talents people pick the most, how often they win with them, what maps they win on, etc. People have pulled out selective dumps trying to confirm or deny long held beliefs like "the team with the lowest MMR player is more likely to lose" and all of that. They've found barely any deviation in those dumps compared to the whole pile of data.

It doesn't "need" to be but it you can't apply the same statistical processes or thinking when it's not. If you have a random sample, you can use a very very small sample size with a very small margin of error as a result. It does not work the same way with a non-random sample. You can still have a non-random sample, and it can still provide information, but it's not as easily and effectively useful as a nonrandom sample. It's going to have inherently larger margins of error.

Ultimately, though, beyond the hotdogs may be/may not be a lie conspiracy theory that we've had going on since Alpha, this is the part of your posts that has flummoxed me.

I'm not, and I don't believe anyone else is talking about looking at the completed talent build section. When I suggested it to the poster, I was referring to the actual majority of the page (http://www.hotslogs.com/Sitewide/HeroDetails) where you can see each individual talent broken down by number of times picked, winrate when picked, and overall popularity for all brackets or for your given bracket or anything in between.

If you're trying to find a build, there's pretty much no more reliable way than to look at that and see that yes, not many people pick Totemic Projection at 7 on Rehgar. As Familie said, you can do with that information what you will, but odds are over a large enough sample size there's a reason certain talents are completely neglected.

Putting aside the strawman conspiracy thing, how did my original post not acknowledge these differences?

The problem with the Hotslog talent data is that it really only tells you what's popular (or due for a nerf). Nearly every talent is going to have a winrate between 45-55% to begin with, and once you factor in a margin of error it's basically a 50/50 wash. Sure, it can help you identify 'trick' talents which you're not supposed to take, or 'OP' talents they haven't tuned yet, but those are few and far between and you could probably figure those out already. Their talent build section is incredibly, incredibly misleading because the win-rate percentages reflected there are not accurate whatsoever. Obviously a hero with an overall win-rate of ~50% and a individual talent win rate of ~50% cannot have 10 talent builds all above 60%, it just doesn't add up. The reason this happens is subtle but obvious once you think about it.

That seems to be a pretty even-handed post which talks about whether Hotslog can provide you with the relevant information you need to make a decision about which talent will help you win. Sure, most people don't pick Totemic Projection, but the win rate is still calculated at 49.3%. Even with a 1% margin of error, that still puts it in the same win range as cleanse even though cleanse is picked 10x more often. Similarly, far-sight has the highest win rate percentage even though it has the second lowest pick percentage.

So if you're a player who wants to know what the best choice at that tier is, what do you do? Even if the win-rate percentages were accurate to within 1%, that would mean that all the choices at that tier would be equivalent and your choice wouldn't seem to matter. It gets worse if you end up with a low-picked talent that has the higher win rate. Xul's level 1 talent Backlash is picked only 28% of the time versus 69% for Shade, but it actually has a 3.2% higher win rate. Does that mean a new player should pick Backlash, or should they go with the herd and pick Shade?

Also, what is inaccurate about how the data is generated again? Uploading only wins or something like that? The sample size is just too large for that man. You can go through every hero and look at the most picked talents, or hell, even two most picked talents on each tier and get near universal agreement on them being the correct choices in most scenarios. What exactly is inaccurate or misleading about that?

What's misleading is the idea that it's providing you any actionable information beyond what's popular. I assume players wants to pick talents that help them win, not just do what's popular. A non-random sample that doesn't disclose or cannot calculate its margin of error isn't useful for making decisions based on win rates. There's also a secondary compounding issue of data generation with respect to level 20 and the full talent build information, but that's not necessary to the broader point I'm trying to make.

 

[Alur]

What's misleading is the idea that it's providing you any useful information beyond what's popular. I assume players wants to pick talents that help them win, not just do what's popular. A non-random sample that doesn't disclose or cannot calculate its margin of error isn't useful for making decisions based on win rates.

What's popular typically = what's also the most useful in most scenarios in the vast majority of these choices is the thing I think we are differing on here. As I've said several times, the majority of these builds shown on hotsdogs are the talents chosen by most people who actually know how to play the hero as well. Whether that's high MMR players, HOTS pros, the talking heads on podcasts, etc. It's not unlike talents in WoW, or anything else. Your implication that it's just "what's popular" is weird based on cherrypicking a few lesser picked talents that have a slightly higher winrate.

A talent that is used in 1/3rd the amount of games having a slightly higher winrate isn't actually indicative of much of anything from my point of view. Through playing the hero and even just taking Xul in try mode you can see that Backlash is at best situationally useful because it's too hard to land the damage. In those situations, you use it and it performs better than Shade. Shade, however, is used in all situations. If Backlash was used 68.7% of the time instead of just 28.1% of the time, I'm fairly confident you'd see a much lower winrate on that talent.

 

[Zackat]

Then why does Blizzard seem to nerf the talents that Hotslogs has identified as the good ones, but others don't get touched or even get buffed? Is it just because they won the popularity contest or are they actually the best talents that have an advantage over the others?

 

[jaundicejuice]

They think Li-Ming is balanced so who knows how they arrive at the conclusions they do.

http://instantrimshot.com/

 

[Alur]

They think Li-Ming is balanced so who knows how they arrive at the conclusions they do.

http://instantrimshot.com/

Haha. And did the same thing with Kael'thas before. Seems like the only heroes that do damage that they rush to quickly correct or worry about are the melee (other than Zeratul).

 

[Brawndo Addict]

What's popular typically = what's also the most useful in most scenarios in the vast majority of these choices is the thing I think we are differing on here. As I've said several times, the majority of these builds shown on hotsdogs are the talents chosen by most people who actually know how to play the hero as well. Whether that's high MMR players, HOTS pros, the talking heads on podcasts, etc. It's not unlike talents in WoW, or anything else. Your implication that it's just "what's popular" is weird based on cherrypicking a few lesser picked talents that have a slightly higher winrate.

A talent that is used in 1/3rd the amount of games having a slightly higher winrate isn't actually indicative of much of anything from my point of view. Through playing the hero and even just taking Xul in try mode you can see that Backlash is at best situationally useful because it's too hard to land the damage. In those situations, you use it and it performs better than Shade. Shade, however, is used in all situations. If Backlash was used 68.7% of the time instead of just 28.1% of the time, I'm fairly confident you'd see a much lower winrate on that talent.

Huh, okay. Yeah, then I guess we just disagree on what the appropriate metric for evaluating talents are I guess. I see win rate as the ideal metric but if you're using pick rate than the inherent error margin problems I'm talking about are going to be completely inconsequential to that analysis.

I'm not sure that position is entirely consistent though, because surely there's a point at which the popular choice is going to be wrong? Like say 75% of people pick X instead of Y, but X has a 40% win rate and Y has a 60% win rate. Surely we would tell people not to pick X?

Plus why does the pick percentage matter if the raw number of picks is still large enough? If 15,000 people pick X and 5,000 pick Y, isn't that enough to estimate Y's win rate? It shouldn't matter that Y was picked only 25% of the time if you only need, say, 3,000 picks to estimate the win rate accurately (not saying that is enough or not enough, just a hypothetical ).

I would differentiate it from WoW talents because you could literally math the dps of builds out using target dummy's or a tank-spank boss. We only have derivative statistics to really base decisions off in MOBAs.

 

[Alur]

I'm not sure that position is entirely consistent though, because surely there's a point at which the popular choice is going to be wrong? Like say 75% of people pick X instead of Y, but X has a 30% win rate and Y has a 70% win rate. Surely we would tell people not to pick X?.

Find me some scenarios where that's the case at Diamond/Master league (which is almost always mirrored by the lower, more played brackets 1-for-1) and I'd be shocked. I don't mean like 2% winrate difference either, I mean a significant gap in winrate like 5% or better.

There's a reason they are popular, just like talent builds in WoW or any MMO are popular, or item builds in League and Dota. They work, by and large, and Blizzard is not particularly adept at creating multiple "good" choices at each tier. Every time we think a hero has diversity, it ends up getting drilled down to the same build with maybe a little variation on a tier or two.

In general the winrate difference on those tiers between two decent (but one often situational) talents is like what you were talking about with Shade and Backlash. 2 or 3% tops and is easily explainable as exploiting the situation you are in with your specific comp and picking situationally as opposed to generally picking the strongest "overall" talent.

I would differentiate it from WoW talents because you could literally math the dps of builds out using target dummy's or a tank-spank boss. We only have derivative statistics to really base decisions off in MOBAs.

Again, you're digging way too deep into this. It's still a talent build. And we do have target dummies actually, they are in try mode. They aren't really needed to tell which talents will shore up the weakness of a hero or give them better utility to enhance their kit, however. And HOTS values certain things very highly and if it's a given heroes talents you can almost guarantee it's the pick (regen, waveclear, cc/slows, etc). Apples and oranges in the way you play the game, but ultimately builds are pretty much cookie cutter across all competitive games.

 

[jaundicejuice]

Haha. And did the same thing with Kael'thas before. Seems like the only heroes that do damage that they rush to quickly correct or worry about are the melee (other than Zeratul).

People who enjoy playing as Li-Ming better savor this time on the top because she's gonna fall hard when they finally admit that she's got to be addressed. They've done multiple revisions to Living Bomb interaction, this is the third major revision of his talent set and they still haven't gotten him right. KT's sort of gotten worse with each revision, not more balanced. This is what awaits Li-Ming.

 

[Brawndo Addict]

Find me some scenarios where that's the case at Diamond/Master league (which is almost always mirrored by the lower, more played brackets 1-for-1) and I'd be shocked. I don't mean like 2% winrate difference either, I mean a significant gap in winrate like 5% or better.

Over what period of time and how many raw picks though? Like does it have to be picked over 1,000 times or only 500? Over a month or 2 weeks? Here's one example going back to 3/20 just for reference.
3/

 

[TDLink]

Did anyone else get a message to restart the launcher for an update and upon doing so the resolution is screwed up?

 

[Alur]

People who enjoy playing as Li-Ming better savor this time on the top because she's gonna fall hard when they finally admit that she's got to be addressed. They've done multiple revisions to Living Bomb interaction, this is the third major revision of his talent set and they still haven't gotten him right. KT's sort of gotten worse with each revision, not more balanced. This is what awaits Li-Ming.

I sadly agree with you on that. Not a good track record when they can't reign a hero in with the first set or two of tweaks. They typically get rekt. Or sometimes they get rekt by skipping the small tweaks. She'll join the long line of heroes before her, like Stitches, Arthas, Brightwing, KT, etc.

I feel like perhaps they should've went harder with the "niche" aspects of each hero.

Want Li-Ming to be super mobile? Reduce her damage by like 25-40%, give her longer range on her blink, and let her talent into even more movement and blink.

Want Jaina to be slow and bursty and have no mobility beyond Ice Block? Make Ice Block base and remove all of her range talents (no more range on Blizzard, etc). Maybe give her some more damage to compensate.

That should leave KT in the middle. He should have less damage than Jaina but more mobility than Jaina. More damage than Li-Ming but less mobility than Li-Ming. Change his damage to be all based on damage over time beyond Pyroblast. Force him to stay at the same range he's currently at, but allow him to talent into movespeed based on dots stacking (a-la the level 1 for Dehaka) or something. Remove Arcane Barrier and give back Bolt.

Over what period of time and how many raw picks though? Like does it have to be picked over 1,000 times or only 500? Over a month or 2 weeks?

To me, you can dismiss anything that's not picked at least half as much as the dominant talent. That's pretty much the definition of niche pick particularly if you're only looking at Diamond and Master because there's only like 10-20k games on most heroes at that level in a 7 day span. Some heroes, like Hammer or TLV or whatever barely have 1500 games played so looking at just Diamond and Masters does skew the numbers a bit.

Like for Diamond and Master, Backlash is picked 21.5% of the time while Shade is picked 77.5% of the time. It's not even close. Backlash has a 6% higher winrate in that scenario, but in most scenarios I would never recommend you take that talent over Shade because it is an absolute lifesaver given your role on the field as Xul.

Pretty much anything you can find can be explained away is my point. There's always outliers, but I doubt you'll find more than a handful that are actually the proper pick the majority of the time yet aren't chosen anyway.

Here's one example going back to 3/20 just for reference.
3/

Nature's Culling is a good one. You're trading damage for wave clear. It's a legit choice. I honestly don't know in that scenario what pros are picking there, but IIRC I've seen people telling Lunaras to go Culling at 7 in game. I think that is what is used competitively but I'm not sure. Someone else would have to chime in, I didn't play a ton of Lunara once they fixed her.

You are actively having to look to find these couple of outliers versus what is popular, in your opinion, which goes against what you were writing those novels about, no? By and large the popular choice is the best choice. I see 7 tiers in that list and one had a discrepancy. Let's say you even found one each on each of the 50 heroes, that'd still be 86% accuracy. That's pretty gat damn accurate for helping you win, which is what your complaint was right?

 

[Ketch]

ITT brawndo addict tries to teach alur how statistics work, and alur doesn't understand.

I think there's also another huge factor you didn't mention which is that all of the data is publicly available and used to determine what talents to picks, which skews the results even more.

 

[Celegus]

Nature's Culling is hilariously better than Vigor. 50% AA damage sounds nice and all, but the majority of Lunara's damage comes from the poison, not the actual attack. With Nature's Culling, she turns into a waveclear machine and bulldozes structures with the best of them. I can see a case for Splintered against something like Morales or Uther, but Culling is what I take 99% of the time. Lunara would probably be my "main" if I really had to pick one.

 

[Alur]

ITT brawndo addict tries to teach alur how statistics work, and alur doesn't understand.

I think there's also another huge factor you didn't mention which is that all of the data is publicly available and used to determine what talents to picks, which skews the results even more.

LOL, I think he clearly said he's not a statistician himself.

The second part is probably true, but that goes the same for any game like this. If it wasn't hotdogs giving us this information, everyone would just use HeroesFire and arrive at the same conclusion more or less. That's just the nature of games requiring builds or decks or anything of that sort. A few people come up with their own stuff, the rest follow the herd after being convinced...what it takes to convince people goes down over time, though, for sure.

 

[kirblar]

Most players aren't religiously looking at HOTSLOGS pickrates and stuff when playing. That's what makes it a relatively good place for data.

 

[brian!]

hotslogs is a good start for thinking out builds dont let it decide ur build tho and there are some egregious things that are picked often

 

[jaundicejuice]

I hope they change's Xul's root to a skill shot or reduce it's duration. It's kind of stupid the way it is.

 

[Alur]

Wonder if we see Dehaka for Dorm this weekend? They are on the live patch IIRC.

 

[brian!]

if he's able to be picked im fairly certain he'll be in there

 

[kirblar]

It's two weeks in, they better be on the live patch.

 

[Syf]

pretty hyped for tastosis tbh

 

[Alur]

Wait, what? Tastosis are casting Dorm again? I didn't see the list, only that Jake got riparoni and pepperoni'd.

 

[jaundicejuice]

I would actually watch my first HotS tourney to see Dehaka play.

 

[Alur]

It's on ESPN2 this time 'round, so you'll need to sign up for the Sling trial to watch it if you don't have cable or the like. Or be an EU resident and get to watch that fancy YouTube stream of the ESPN2 broadcast.

 

[Brawndo Addict]

Okay last derail I promise.

Like for Diamond and Master, Backlash is picked 21.5% of the time while Shade is picked 77.5% of the time. It's not even close. Backlash has a 6% higher winrate in that scenario, but in most scenarios I would never recommend you take that talent over Shade because it is an absolute lifesaver given your role on the field as Xul.

Pretty much anything you can find can be explained away is my point. There's always outliers, but I doubt you'll find more than a handful that are actually the proper pick the majority of the time yet aren't chosen anyway.

You are actively having to look to find these couple of outliers versus what is popular, in your opinion, which goes against what you were writing those novels about, no? By and large the popular choice is the best choice. I see 7 tiers in that list and one had a discrepancy. Let's say you even found one each on each of the 50 heroes, that'd still be 86% accuracy. That's pretty gat damn accurate for helping you win, which is what your complaint was right?

I'm not sure how Xul doesn't prove my exact point. He also has the level 10 disparity where Nova is picked 65% and Mages is picked 35% but mages has the higher win rate by about 5%. The crowd pick is statistically worse than the minority pick according to that data set. That should lead us to conclude that people over-value the popular choice even though the alternative pick is more likely to win.

My point is that popularity is not as useful of a metric for determining how good a talent is as the win rate is. We value heroes based on their win percentages first and foremost after all, not their pick percentages, I don't see why talents should be any different. A talent or hero being popular doesn't make it more likely to win the game. A talent or hero being more likely to win the game makes it more likely to be popular. The vast majority of the time the two things are going to be aligned because people like to win, but not always. If you could only have one statistic about which ultimate to pick, would you rather know which one was picked the most, or which one won the most?

If we can't distinguish between a "niche" or "explainable" pick versus a truly under-valued pick, then we've basically set up a scenario where the popular choice is completely unfalsifiable. A minority pick with a higher win rate will always be "explained" away until popular opinion shifts and the pick percentages reverse. You see that all the time in MOBAs, no balance changes actually happen, but suddenly a different item/build becomes super popular that used to be considered bad, usually spurned on by a player or team that consistently makes the 'bad' choice.

Obviously I'd still argue that, once you introduce a margin of error, most talent tiers are going to be equivalent in terms of win-rate ranges. Outperforming talents are usually indicative of tuning/balancing issues, see Xul/Li-Ming.

LOL, I think he clearly said he's not a statistician himself.

I'm not a statistician but the topic here isn't particularly nuanced, it's more high-school/college stats. If I had to talk about how to weight data from games that reach lvl 20 versus games that don't reach level 20, then I would be in some serious trouble.

 

[Maledict]

I'm not sure how much of this is silly rumour mill stuff, but apparently Erho is out of Naventic and Srey out of Tempostorm?

Edit: Erho confirmed, and apparently it was because of his attitude. See I knew I was right when I said he came across as a dick on Townhall Heroes and his stream!

He's duoing with Srey right now, so god knows what is going on.. Don't know who replaces Erho in Naventic, or how Tempo fill the two spots they have open now. Rosterpocalypse has hit NA!

EDIT2: Weird how all the NA pros outside of C9 are suddenly playing hero league. do you think they came back from Korea and realised that they actually need to be doing it for practice?

 

[Familienoberhauptvogel]

Well they are still stuck with Kenma tho. Unless CoG also does a shuffle this means it's their and Blaze's best chance to qualify for the foreseeable future. Except the regional qualifier isn't for a while.

 

[SRTtoZ]

So....I'm all of a sudden addicted to this game, even though I always thought MOBAS were shit? I'm sitting here getting ready to play my recently purchased copy of Legacy of the Void, the FINAL chapter in the Starcraft 2 SAGA!, and while looking at the Blizzard Launcher I notice HOTS is the only game I don't have installed and my OCD got the best of me so I said fuck it and installed it.

After it was done, I was just wanted to check it out quick to see what it looked like. I knew it was a MOBA, I knew it was F2P, earned coins, all that bullshit, and I even knew the IGN meme 6.5/10 no comeback mechanics. Anyway, so fast forward 4 days later and I have 6 characters leveled past 5, some even higher, I bought the starter pack just to have something, I have Artanis and Valla for some reason. I'm sitting on like 20k gold after buying Reghar just now and I'm fucking hooked. Blizzard games have always been my favorite but I never saw a hype mechanics behind Mobas but I finally see the light. Damn I got a new favorite game.

Oh forgot to mention, I always knew people said Mobas are damn toxic but man, when you're losing in this game people get fucking PISSED. Every game ends with someone saying 'GG CARRIED' or some bullshit. Whats up with that? Sometimes I type while I'm dead to try and communicate with people and I get yelled at. That's the only downside so far, but whatever I'm just trying to get to know all the heroes so I can play Ranked and get out of this QM bullshit. Any advice is appreciated, I love to play support but Illidan's hunt build is fucking rad.

 

[Milly79]

Lol. I hope Erho just leaves the scene. He's been on 2-3 teams now and they've all stated his attitude fucking blows. Not hard for us to figure out either.

And Srey, eh? Not really too fond of him either.