Page 35 of 118 FirstFirst ...
25
33
34
35
36
37
45
85
... LastLast
  1. #681
    Dreadlord zenga's Avatar
    Join Date
    Jun 2010
    Location
    Belgium
    Posts
    987
    GC just posted on the EU forums an answer to a destruction warlock question for a single target buff. Guess this applies to elemental shamans as well:

    Q:Appreciate the update Ghostcrawler, but still waiting on news of a buff for Destro single target damage.

    A: As I said above, it makes more sense for us to implement mechanics changes before we tune all the numbers. Otherwise, you throw out all the tuning work every time you make another mechanics change. I will post when we've finished our 5.2 tuning pass so you guys have a chance to offer feedback.

    We think Destro may be a little behind, but it's not much. Affliction PvE damage is too high, which causes many warlocks to go Affliction on every fight rather than considering options. That creates a big sampling bias effect on sites like World of Logs and Raidbots. The lower performing DPS specs appear even lower than they actually are because the only players generating those logs are uninformed (which also tends to mean undergeared) or just messing around. To get a more accurate test, you need to see a player try every spec under similar circumstances (gear, skill, etc.), and those data are much more rare.


    The 2nd part doesn't really apply to ele as we can't switch to enhance without gear, whereas pures can.

  2. #682
    Legendary! Raiju's Avatar
    Join Date
    Apr 2011
    Location
    England
    Posts
    6,156
    Yes, using all parses makes so much sense. That's only logical when there aren't a huge amount of parses to begin with (to make sure the sample size is reasonable..) Why would you use that when you have top 100 and months of data to have plenty of logs?

    Inb4 "but it's only 100 parses" it's the top 100. It's the top 100 for a reason. So what, a more FOTM spec gets lower on all parses because FOTM players are naturally bad?

  3. #683
    Moderator Endus's Avatar
    Join Date
    Feb 2010
    Location
    Waterloo, ON
    Posts
    21,389
    Quote Originally Posted by zenga View Post
    The 2nd part doesn't really apply to ele as we can't switch to enhance without gear, whereas pures can.
    It does, somewhat; a lot of Shaman go Enhancement because Elemental is performing so much worse. It's not as easy to switch, so if a patch suddenly hosed Enhancement and boosted Elemental you wouldn't see as fast a turnaround, but it has a long-term effect.

    That's just why it'll take a few weeks/months to see Elemental's true recovery once they're fixed, though, as people gear up their Elemental spec again, not an explanation for their current numbers.

    Quote Originally Posted by Raiju View Post
    Yes, using all parses makes so much sense. That's only logical when there aren't a huge amount of parses to begin with (to make sure the sample size is reasonable..) Why would you use that when you have top 100 and months of data to have plenty of logs?

    Inb4 "but it's only 100 parses" it's the top 100. It's the top 100 for a reason. So what, a more FOTM spec gets lower on all parses because FOTM players are naturally bad?
    Because "top X parses" is statistically dishonest and misleading for a host of reasons, always, regardless of the amount of data. In fact, as you get more data, it makes "top X parses" models MORE misleading, not less.

    I dealt with this in excruciating detail in a blog post a while back. The specific numbers are obviously out of date, but the math theory that is the point of the post is math theory, and never goes out of date; http://www.mmo-champion.com/threads/880800
    Last edited by Endus; 2013-01-24 at 06:31 PM.

  4. #684
    Quote Originally Posted by zenga View Post
    GC just posted on the EU forums an answer to a destruction warlock question for a single target buff. Guess this applies to elemental shamans as well:

    Q:Appreciate the update Ghostcrawler, but still waiting on news of a buff for Destro single target damage.

    A: As I said above, it makes more sense for us to implement mechanics changes before we tune all the numbers. Otherwise, you throw out all the tuning work every time you make another mechanics change. I will post when we've finished our 5.2 tuning pass so you guys have a chance to offer feedback.

    We think Destro may be a little behind, but it's not much. Affliction PvE damage is too high, which causes many warlocks to go Affliction on every fight rather than considering options. That creates a big sampling bias effect on sites like World of Logs and Raidbots. The lower performing DPS specs appear even lower than they actually are because the only players generating those logs are uninformed (which also tends to mean undergeared) or just messing around. To get a more accurate test, you need to see a player try every spec under similar circumstances (gear, skill, etc.), and those data are much more rare.


    The 2nd part doesn't really apply to ele as we can't switch to enhance without gear, whereas pures can.
    I think the bold text is what is interesting for us

  5. #685
    Quote Originally Posted by Raiju View Post
    Yes, using all parses makes so much sense. That's only logical when there aren't a huge amount of parses to begin with (to make sure the sample size is reasonable..) Why would you use that when you have top 100 and months of data to have plenty of logs?

    Inb4 "but it's only 100 parses" it's the top 100. It's the top 100 for a reason. So what, a more FOTM spec gets lower on all parses because FOTM players are naturally bad?
    Using the best 100 parses from every spec is completely arbitrary and entirely meaningless as a comparative metric, because population between specs isn't even close to balanced.

    Raidbots has a sample pool of 21,393 Fury Warrior parses but only 8,816 for Enhancement Shamans.

  6. #686
    Moderator Endus's Avatar
    Join Date
    Feb 2010
    Location
    Waterloo, ON
    Posts
    21,389
    Quote Originally Posted by Nitwit View Post
    Using the best 100 parses from every spec is completely arbitrary and entirely meaningless as a comparative metric, because population between specs isn't even close to balanced.

    Raidbots has a sample pool of 21,393 Fury Warrior parses but only 8,816 for Enhancement Shamans.
    Plus different standard deviations between specs.
    Plus fight-specific shenanigans being overvalued.
    Etc.

    It amounts to deliberate sample bias. There's literally no reason to use top X models for class balance discussions.

  7. #687
    Quote Originally Posted by zenga View Post
    GC just posted on the EU forums an answer to a destruction warlock question for a single target buff. Guess this applies to elemental shamans as well:

    Q:Appreciate the update Ghostcrawler, but still waiting on news of a buff for Destro single target damage.

    A: As I said above, it makes more sense for us to implement mechanics changes before we tune all the numbers. Otherwise, you throw out all the tuning work every time you make another mechanics change. I will post when we've finished our 5.2 tuning pass so you guys have a chance to offer feedback.

    We think Destro may be a little behind, but it's not much. Affliction PvE damage is too high, which causes many warlocks to go Affliction on every fight rather than considering options. That creates a big sampling bias effect on sites like World of Logs and Raidbots. The lower performing DPS specs appear even lower than they actually are because the only players generating those logs are uninformed (which also tends to mean undergeared) or just messing around. To get a more accurate test, you need to see a player try every spec under similar circumstances (gear, skill, etc.), and those data are much more rare.


    The 2nd part doesn't really apply to ele as we can't switch to enhance without gear, whereas pures can.
    Destruction is also great, but not quite as overpwoered as affli.

    Look here:
    http://raidbots.com/dpsbot/Spec_Scor.../7/30/default/

    All three warlock speccs are top, followed by arcane and rogues.

  8. #688
    Quote Originally Posted by Endus View Post

    It amounts to deliberate sample bias. There's literally no reason to use top X models for class balance discussions.
    Even Patchwerk style sim rankings on Simcraft?

  9. #689
    High Overlord Etherius's Avatar
    Join Date
    Jan 2013
    Location
    British Columbia
    Posts
    103
    I'm looking forward to these changes~
    Ever since TBC I've been playing my ele shaman, that as far as I'm concerned, will never change ^_^

  10. #690
    Quote Originally Posted by Endus View Post
    It does, somewhat; a lot of Shaman go Enhancement because Elemental is performing so much worse. It's not as easy to switch, so if a patch suddenly hosed Enhancement and boosted Elemental you wouldn't see as fast a turnaround, but it has a long-term effect.

    That's just why it'll take a few weeks/months to see Elemental's true recovery once they're fixed, though, as people gear up their Elemental spec again, not an explanation for their current numbers.



    Because "top X parses" is statistically dishonest and misleading for a host of reasons, always, regardless of the amount of data. In fact, as you get more data, it makes "top X parses" models MORE misleading, not less.

    I dealt with this in excruciating detail in a blog post a while back. The specific numbers are obviously out of date, but the math theory that is the point of the post is math theory, and never goes out of date; http://www.mmo-champion.com/threads/880800
    There are still more Elemental than Enhancement shamans. There's not a lot of change, though enhancement got a little closer.

    12 months: Ele to Enh: 1.11
    6 months: 1.5
    3 months: 1.63
    1 month: 1.51
    now: 1.42

    But it's not like they simply respecc, reforge and suddenly can deal competitive dps.

  11. #691
    Moderator Endus's Avatar
    Join Date
    Feb 2010
    Location
    Waterloo, ON
    Posts
    21,389
    Quote Originally Posted by Thanntos View Post
    Even Patchwerk style sim rankings on Simcraft?
    Sims are inherently NOT "top X" models. They run X iterations (usually 10k), and determine the averages based on all 10k of them.

    They don't use the top 100 of those simulated parses for the same reasons top X models with actual logs are awful; because it's just deliberately biasing the data for no good purpose.

  12. #692
    The lower performing DPS specs appear even lower than they actually are because the only players generating those logs are uninformed (which also tends to mean undergeared) or just messing around.
    Am I the only one slightly offended by this statement? It feels like he's saying Elemental is only behind because everyone who plays it currently is substandard in skill and/or messing around...

  13. #693
    Legendary! Raiju's Avatar
    Join Date
    Apr 2011
    Location
    England
    Posts
    6,156
    I did statistics, and your little post means absolutely nothing to people who frequently hit top 100's (the ranks that would be involved here). Random variation? Not multiple kills in a night, multiple nights in a week/month, etc. You don't get so many people who consistently get the high ranks because they're lucky.
    History? It can apply, but as far as I'm aware nothing hit live recently.
    Player bias, given there were no rogues or warlocks in most of my runs (and a frost dk only if unholy can still give frenzy to others) I don't see how this could apply either. This doesn't apply also when multiple in the same run get ranks on the same kills consistently.

    Player skill (and gear) is huge. Now at this point most of the people who are raiding frequently are geared. That means that this will have little effect on the results, although naturally some (what maybe we count 200 dps either way?)

    Standard deviation obviously can't be ignored, but then having played a warrior a lot of the damage is controlled. Much like enhance in 5.X. Much like elemental. We aren't fire mages, if anything all 3 specs are closer to cata arcane mages. Still taking the top percentile of a large amount of players (yes 8k of logs is good for a 100 sample...) you are going to run in issues with specs that are significantly more RNG, surely this just runs into random variation though?

    But then, half the people here would refute with spec being fine by using simcraft.. says it all about the general level of analysis around these days.

  14. #694
    Moderator Endus's Avatar
    Join Date
    Feb 2010
    Location
    Waterloo, ON
    Posts
    21,389
    Quote Originally Posted by Kinnetik View Post
    Am I the only one slightly offended by this statement? It feels like he's saying Elemental is only behind because everyone who plays it currently is substandard in skill and/or messing around...
    It's referring more to the specs that are so low they don't have enough parses to generate reasonable stats, like Arms warriors. There's a certain percentage of "just faffing about to test some stuff" in the parses, for everyone, and for those specs, the faffing about is most of the parses in question.

    ---------- Post added 2013-01-24 at 02:31 PM ----------

    Quote Originally Posted by Raiju View Post
    I did statistics
    If you really had, and you mean something more than a course you took in high school, then you should have been taught about how to bias a sample.

    Deliberately selecting a sample based on certain factors always produces bias. There's no reason to exclude data. Top X parse models are inferior in pretty much every imaginable way, and as I went through years ago in that thread, the differences we're talking about in standard deviations and such absolutely ARE significant enough to sway the numbers hugely.

    We're already looking at successful Heroic kills. That already controls for skill, enough to eliminate almost all the really bad players. You don't need to trim it further.

  15. #695
    Dreadlord zenga's Avatar
    Join Date
    Jun 2010
    Location
    Belgium
    Posts
    987
    Quote Originally Posted by Endus View Post
    It does, somewhat; a lot of Shaman go Enhancement because Elemental is performing so much worse. It's not as easy to switch, so if a patch suddenly hosed Enhancement and boosted Elemental you wouldn't see as fast a turnaround, but it has a long-term effect.

    That's just why it'll take a few weeks/months to see Elemental's true recovery once they're fixed, though, as people gear up their Elemental spec again, not an explanation for their current numbers.
    That's a possibility of course, but it remains speculation. Personally I don't know anyone who at the start of MoP wanted to play ele but figured they were gonna be shit and went enhancement instead. From my experience it's the contrary: most ele shamans I know play their mage/shadow priest/warlock alt at a much higher standard than they play enhancement (and some of them just rerolled to one of these specs). But if that amount of players is big enough to have an influence on the overall pictures, ... I have my reservations. Contrary to mages for example, where frost and fire dropped massively when all mages went arcane. Point being: for pures that rule of thumb is more or less established, for hybrids not at all. At least not to the extend to draw conclusions.

  16. #696
    Legendary! Raiju's Avatar
    Join Date
    Apr 2011
    Location
    England
    Posts
    6,156
    Quote Originally Posted by Endus View Post
    We're already looking at successful Heroic kills. That already controls for skill, enough to eliminate almost all the really bad players. You don't need to trim it further.
    This is what we call an opinion displayed as fact, ladies and gentlemen.

  17. #697
    Quote Originally Posted by Raiju View Post
    This is what we call an opinion displayed as fact, ladies and gentlemen.
    No, that's just plainly his opinion. However as Endus is well known here on MMO-C and been helping with lots of shaman related stuff many people take his opinions as very close to facts. You are free to challenge his opinion, but like it's been proven your solution got quite a bit of loopholes in it. Also adding the "ladies and gentlemen" statement in the end just makes you sound like a total douchebag, and it won't make people take your side most of the time.

  18. #698
    I wonder, anyone have an idea about what is the reasoning behind the recent "buff" to enhance?
    "Mental Quickness now has the additional bonus of making your shock spells cost 90% less mana."
    I just wonder, it is already at 75% less mana on shocks, so why is it getting even lower? Like we really need those few hundred mana

  19. #699
    Moderator Endus's Avatar
    Join Date
    Feb 2010
    Location
    Waterloo, ON
    Posts
    21,389
    Quote Originally Posted by Telefonorm View Post
    No, that's just plainly his opinion.
    Well, specifically, it's an argument, backed up by the math theory that I posted from that post I wrote years back that I didn't bother to repeat.

    However as Endus is well known here on MMO-C and been helping with lots of shaman related stuff many people take his opinions as very close to facts. You are free to challenge his opinion, but like it's been proven your solution got quite a bit of loopholes in it. Also adding the "ladies and gentlemen" statement in the end just makes you sound like a total douchebag, and it won't make people take your side most of the time.
    And while I support people challenging my arguments, you'd better be bringing equal or greater artillery to bear, to demonstrate where my argument was flawed. People generally trust me now, but at the time I wrote that post, most people thought I was being a jerk and was wrong. This particular argument I'm pretty darned confident in, because almost everyone took a shot at it, back when I wrote it. I was in the minority then, and today, almost everyone uses "all parses", for the reasons I detailed in that post.

  20. #700
    So, what is the chance that Blizzard will un-link the cooldowns of shock spells?

    Better or worse than the chance that they'll give Lava Burst the same crit scaling mechanic as Chaos Bolt?

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •