No… that’s not what a “rotation” is. The term refers to a sequence of offensive abilities.
Fair enough; I always thought that “rotation” included any abilities, both offensive and defensive, but your terminology works too. That said, if you have an ability that builds up aggro faster than the tank can compensate for it, then you will end up either casting FD periodically—and while you’re casting FD, you’re not casting Steady Shot. This is all that I meant. That said, I was unaware that FD was off the GCD; I don’t remember if it was always like that or if they changed it at some point.
I think you’re being somewhat pedantic, and missing my point.
Your original post made it sound as though you only cared about the final DPS readout, to the exclusion of anything else, such as aggro management. When I still played WoW, there were plenty of players who thought exactly that way; you couldn’t throw a snowball without hitting one. These people were the leading cause of death in dungeons. I understand that you are not one of them, but I think I can be forgiven for misinterpreting your original post, given the prior.
That said, in the more general sense the DPS aggro monkeys were victims of over-fitting; and my point was that declaring actions “worse than useless” is often—though obviously not always—a symptom of over-fitting, or perhaps merely of solving different goals. For example, the fencer who ends every move with a flourish might have a very different goal in mind (“showing off”, “roleplaying”, “added challenge”, etc.) than a purely competitive player whose only goal is “winning the match as quickly as possible”.
From the point of view of the competitive player, his opponent is indeed performing “worse than worthless” actions; yet the assessment is still incorrect.
From your description of events in WoW, I understand that the above probably does not apply to those Hunters whom you’d described; however, I still maintain that it is often the case.
That said, if you have an ability that builds up aggro faster than the tank can compensate for it, then you will end up either casting FD periodically—and while you’re casting FD, you’re not casting Steady Shot.
This is true, and in fact I have an amusing anecdote about this very fact (amusing in retrospect, of course).
I understand that you are not one of them [people who only care about DPS and not threat], but I think I can be forgiven for misinterpreting your original post, given the prior.
Granted. The prior is pretty strong there, true enough.
That said, in the more general sense the DPS aggro monkeys were victims of over-fitting
I’m not entirely sure what you mean here. Could you clarify?
From your description of events in WoW, I understand that the above probably does not apply to those Hunters whom you’d described; however, I still maintain that it is often the case.
Well, perhaps. I don’t have anywhere near as strong an opinion on this once we broaden the context beyond games (even to games I’m unfamiliar with); you may well be right. I wonder if we couldn’t try and break down the categories of activity in which apparently “worse than worthless” things can be reasonably construed as optimizations for different targets?
A couple of preliminary stabs at an answer:
Whether the person understands that, and acknowledges that, their behavior is not optimized for the default target, should affect our probability estimate of whether the behavior is just a failed attempt at optimization, or successful optimization for something else.
For our purposes, there’s a difference between activities or aspects of activities that have a performance component and ones that do not; e.g. fencing (people are watching you) vs. WoW huntering (no one can see what buttons your are pressing on your keyboard).
That said, in the more general sense the DPS aggro monkeys were victims of over-fitting
I’m not entirely sure what you mean here. Could you clarify?
Right, sorry, that was kind of a weird metaphor on my part.
Imagine that you’re building a classifier (f.ex. a neural network or something) that tells you which houses are “good” and which are “bad”. You have a bunch of features that describe each house; among them are longtitude and latitude. You feed some training data to the classifier, hoping that it will draw you some volume in the feature-space. Everything inside the volume is “good”, everything outside of it is “bad”.
Over-fitting occurs when you don’t give the classifier enough training data, and when you punish it too severely for any errors. Thus, it either ends up drawing a really convoluted shape that fits exactly those 5 data points that you gave it; or it ends up ignoring most of the features because they didn’t have as much of an impact on the outcome; or both. So, your classifier gets a perfect score on the 10 houses it knows about, and fails miserably at real-world data.
Similarly, these Hunters, and many WoW players in general, have been trained on low-level encounters, where the mobs are weak and the penalty for dying is small. They don’t know anything about aggro management, mana conservation, bandaging, etc… All they ever look at is the DPS readout. Thus, when they get to end-game content, they fail miserably.
You can sometimes see the same pattern occur in other situations, too. For example, programmers who are used to working with microcontrollers typically end up writing really terrible host-computer code. Their code is extremely fast and uses very little memory… and is also impossible to modify or maintain. In a microcontroller, using 2x less RAM is the difference between success and failure; on a PC, using 2x less RAM is usually not even noticeable.
e.g. fencing (people are watching you) vs. WoW huntering (no one can see what buttons your are pressing on your keyboard).
You might be right about that, though every WoW move does have a corresponding animation that other people (as well as yourself) can see. I know at least one person who picks their characters based on how awesome they look, but I have no idea whether that is a common pattern or not.
Fair enough; I always thought that “rotation” included any abilities, both offensive and defensive, but your terminology works too. That said, if you have an ability that builds up aggro faster than the tank can compensate for it, then you will end up either casting FD periodically—and while you’re casting FD, you’re not casting Steady Shot. This is all that I meant. That said, I was unaware that FD was off the GCD; I don’t remember if it was always like that or if they changed it at some point.
Your original post made it sound as though you only cared about the final DPS readout, to the exclusion of anything else, such as aggro management. When I still played WoW, there were plenty of players who thought exactly that way; you couldn’t throw a snowball without hitting one. These people were the leading cause of death in dungeons. I understand that you are not one of them, but I think I can be forgiven for misinterpreting your original post, given the prior.
That said, in the more general sense the DPS aggro monkeys were victims of over-fitting; and my point was that declaring actions “worse than useless” is often—though obviously not always—a symptom of over-fitting, or perhaps merely of solving different goals. For example, the fencer who ends every move with a flourish might have a very different goal in mind (“showing off”, “roleplaying”, “added challenge”, etc.) than a purely competitive player whose only goal is “winning the match as quickly as possible”.
From the point of view of the competitive player, his opponent is indeed performing “worse than worthless” actions; yet the assessment is still incorrect.
From your description of events in WoW, I understand that the above probably does not apply to those Hunters whom you’d described; however, I still maintain that it is often the case.
This is true, and in fact I have an amusing anecdote about this very fact (amusing in retrospect, of course).
Granted. The prior is pretty strong there, true enough.
I’m not entirely sure what you mean here. Could you clarify?
Well, perhaps. I don’t have anywhere near as strong an opinion on this once we broaden the context beyond games (even to games I’m unfamiliar with); you may well be right. I wonder if we couldn’t try and break down the categories of activity in which apparently “worse than worthless” things can be reasonably construed as optimizations for different targets?
A couple of preliminary stabs at an answer:
Whether the person understands that, and acknowledges that, their behavior is not optimized for the default target, should affect our probability estimate of whether the behavior is just a failed attempt at optimization, or successful optimization for something else.
For our purposes, there’s a difference between activities or aspects of activities that have a performance component and ones that do not; e.g. fencing (people are watching you) vs. WoW huntering (no one can see what buttons your are pressing on your keyboard).
Right, sorry, that was kind of a weird metaphor on my part.
Imagine that you’re building a classifier (f.ex. a neural network or something) that tells you which houses are “good” and which are “bad”. You have a bunch of features that describe each house; among them are longtitude and latitude. You feed some training data to the classifier, hoping that it will draw you some volume in the feature-space. Everything inside the volume is “good”, everything outside of it is “bad”.
Over-fitting occurs when you don’t give the classifier enough training data, and when you punish it too severely for any errors. Thus, it either ends up drawing a really convoluted shape that fits exactly those 5 data points that you gave it; or it ends up ignoring most of the features because they didn’t have as much of an impact on the outcome; or both. So, your classifier gets a perfect score on the 10 houses it knows about, and fails miserably at real-world data.
Similarly, these Hunters, and many WoW players in general, have been trained on low-level encounters, where the mobs are weak and the penalty for dying is small. They don’t know anything about aggro management, mana conservation, bandaging, etc… All they ever look at is the DPS readout. Thus, when they get to end-game content, they fail miserably.
You can sometimes see the same pattern occur in other situations, too. For example, programmers who are used to working with microcontrollers typically end up writing really terrible host-computer code. Their code is extremely fast and uses very little memory… and is also impossible to modify or maintain. In a microcontroller, using 2x less RAM is the difference between success and failure; on a PC, using 2x less RAM is usually not even noticeable.
You might be right about that, though every WoW move does have a corresponding animation that other people (as well as yourself) can see. I know at least one person who picks their characters based on how awesome they look, but I have no idea whether that is a common pattern or not.