Guild Wars 2

You know what Nike, when I read your post I thought hell yeah that would be cool! But then I told myself wait, if we get to calculating a “difficulty scale” for each boss then maybe we have faced them a bit too much and it is high time we get new content

Wethospu.6437:

I don’t understand either why this is really needed. On static target math should get the same result. If it doesn’t then your math sucks and you should work on that. Once you know that your math works you can generate any kind of situation and try to simulate reality.

I understand wanting check if animations are timed correctly or that skills actually do what they say. Like calculate average DPS for Axe chain and then just Axe chain the Ooze to death. And see if they match.

Because the game is played empirically.

I don’t understand either why this is really needed. On static target math should get the same result. If it doesn’t then your math sucks and you should work on that. Once you know that your math works you can generate any kind of situation and try to simulate reality.

I understand wanting check if animations are timed correctly or that skills actually do what they say. Like calculate average DPS for Axe chain and then just Axe chain the Ooze to death. And see if they match.

So,
A System is actually a pretty complicated thing.

Just to give you some general idea of how complicated, look at chess.
You can look at each and every piece, and easily understand how it impacts the board. But once you start analyzing how those pieces can possibly interact with eachother during a game, the outcomes skyrocket beyond that initial simplicity. There are more possible games of chess than there are atoms in the known universe. That’s a number with over 120 zeroes at the end.

Of course, that’s just a cute little example that includes things like obviously stupid moves. But it illustrates the beast we’re dealing with here in terms of computational complexity, and why scrutinizing the pieces just doesn’t get you much. You could stare all day at a rook, but it isn’t going to make you any better at chess.

In any serious kind of discussion like this you wouldn’t just look at what’s possible, you’d look at what’s probable to narrow things down a bit. And the best way to get some understanding of how things will Probably pan out in practice, is watching to see how it pans out in practice.

/edit: clarity

Damage formula is extremely simple and isn’t exactly same thing as chess. But I will play along. Static boss so we assume that opponent doesn’t move his pieces. Figuring out optimal tactic isn’t that complicated anymore.

Broadicea.8294:

Because the game is played empirically.

Damage formula has already been tested empirically. Not sure what else you would test on a static boss (expect animations and that kind of stuff).

Actually, to me, the most interesting thing is how dynamic combat changes what is optimal in game versus paper.

Arah P1 Record, for example: 3w/1m/1t.

Max dps group comp on paper: 1w/1m/1r/1ele/1eng

If the game translated from spreadsheets to actual game play so well, the spreadsheet comp above would hold every record.

Static meatbag target is not dynamic.

Try not to conflate a chess ‘move’ with actual movement in GW2.
In GW2 using skills would count towards the possibilities.

The closer you get to testing in something to the conditions you would apply that knowledge in, the better. So something played by 5 actual people instead of a spreadsheet is better. Something played by 5 actual people against a mob attacking you back is better than something not attacking you back. Something played by 5 actual people against a mob attacking you back and moving would be even better than that. Etc.

…but there’s limits to what a human being can compute, much less a ragtag group of pseudo-scientists with way too much time on our hands. So we sort of have to go a bit abstract with it just to make it possible, but the closer you get to actual gameplay conditions the better it will be.

For example; people wanted to get better at Chess.
But they knew there was no way they were going to conquer a possibility space larger than the known universe. So they set about to remove possibilities until they reached a point they could manage the rest, while still having conditions they felt were somewhat likely to happen. So they removed all but 6 pieces from the board, and Partially Solved it, trying to learn as much as they could from that. They could’ve just left two pieces on the board and that would’ve been something they could compute, but what good is that to anybody?

There, you said it by yourself.

What’s point of playing with 2 pieces? What’s point of testing with static target?

Wethospu.6437:

Static meatbag target is not dynamic.

I’m not advocating for that. I’m personally satisfied the dps numbers I calculate translate to gameplay closely enough to be meaningful. Like I said, I am much more intrigued by the times when numbers don’t match experience, because that discrepancy is where refinement of skill and innovative strategies pay the most dividends.

Hm. I think my chess metaphors might be getting a bit confusing.
Let me try that again.

A meatbag target is better than a test dummy target because there are actual people playing it for roughly the period of time a boss battle takes. Even at the sub-dodge level of danger, it adds things to the experience nobody can calculate at the spreadsheet level. Like, the performance variance you might get between focused attention span versus sustained attention span.

Yes, it’s an abstract watered-down situation.
Yes, it would be better if they selected more realistic targets. Personally I’d like to see Golem Boss over Ooze. But it’s not an all or nothing kind of thing, nobody walked away from the 6-piece computation with a foolproof winning strategy to chess, but they learned alot about the game in the process. You don’t need to actually find optimal in order for the pursuit of it to be a worthwhile contribution.

Sure, in real situation you will always do worse than optimally. However, pretty much every build is so simple to play that difference between optimal and actual should remain same.

You’re right in thinking Skill Level has an impact.
But that’s not the alpha and omega of why things are different between spreadsheet and practice.

Any moderately complex system doesn’t predictably paint by-the-numbers once things are in motion (weather). Much less a moderately complex system with any kind of human input (The stock exchange). Even systems that are constructed by man with seemingly limited ways for a human to make input and theoretically limited variation (traffic).

It’s not that you’re using math versus not using math.
(If anything running a field test is just giving you more to math with.)
It’s that to learn more about what’s going in math theory alone you’d need much much much higher level math than the spreadsheet level. And, frankly, nobody has time for that. So it’s just much easier to learn more about what’s going on by running a field test.

/edit: last paragraph was a bit oddly written.

I have a lot of time and I love maths

By all means, go for it. :p

I think the point I am making is…

On paper 1w/1r/1eng/1ele/1m is the highest dps party comp in the game give or take. Yet in most cases, speed runs are faster with 4w/1*.

The difference in times between the paper dps group and the real world is what intrigues me. Obviously the answer is shouts are instant cast and can be done on the move, and warriors are the only class besides engie that are good at might stacking AND vuln stacking. And unlike engie they can do it on the run. But nonetheless, I do like exploring that gap in paper and reality.

Nike Porphyrogenita.8137:

I think the point I am making is…

On paper 1w/1r/1eng/1ele/1m is the highest dps party comp in the game give or take. Yet in most cases, speed runs are faster with 4w/1*.

The difference in times between the paper dps group and the real world is what intrigues me. Obviously the answer is shouts are instant cast and can be done on the move, and warriors are the only class besides engie that are good at might stacking AND vuln stacking. And unlike engie they can do it on the run. But nonetheless, I do like exploring that gap in paper and reality.

Agreed. This is one reason I want to attempt to empirically verify the individual dps contributions.

Vox Hollow.2736:

You’re right in thinking Skill Level has an impact.
But that’s not the alpha and omega of why things are different between spreadsheet and practice.

Any moderately complex system doesn’t predictably paint by-the-numbers once things are in motion (weather). Much less a moderately complex system with any kind of human input (The stock exchange). Even systems that are constructed by man with seemingly limited ways for a human to make input and theoretically limited variation (traffic).

It’s not that you’re using math versus not using math.
(If anything running a field test is just giving you more to math with.)
It’s that to learn more about what’s going in math theory alone you’d need much much much higher level math than the spreadsheet level. And, frankly, nobody has time for that. So it’s just much easier to learn more about what’s going on by running a field test.

/edit: last paragraph was a bit oddly written.

Are you seriously comparing systems like weather and traffic to a static meatbag target in a game (Ancient Ooze)?
Could you tell me what’s so complicated about it? It doesn’t attack (well, does, but can’t hit you) so you just have to stand still and do your DPS rotation. Which is exactly what happens in spreadsheet.

I’m trying to present a pretty complicated concept in the easiest way I can imagine. Everybody can identify with weather and traffic, so they make good examples.

Well, there’s the human factor.
Spreadsheets don’t lapse into autopilot-level attention after 8 seconds and clip a back-loaded autoattack with a utility skill, for example. They don’t lose track of time and swap to the second weapon-set too early because they can’t see the cooldown on the screen. They don’t have brains that are constantly taking the easy way out, or fingers with a built-in lag, or eyes that only look at one thing at a time. It’s not just that spreadsheets are perfect, it’s that they’re not human.

Although, I’ll say it again. The closer you get to actual combat conditions the better, so personally I’d like to see a field test go a bit farther than Ooze so you can learn more.

Rotations arent hard to keep on track of. The things that alter it in real combat is situational skills and utility use and dodging attacks. But thats only a small amount of deviation from the optimum dps and really doesnt require that much complicated maths to factor in.

Vox Hollow.2736:

I’m trying to present a pretty complicated concept in the easiest way I can imagine. Everybody can identify with weather and traffic, so they make good examples.

Well, there’s the human factor.
Spreadsheets don’t lapse into autopilot-level attention after 8 seconds and clip a back-loaded autoattack with a utility skill, for example. They don’t lose track of time and swap to the second weapon-set too early because they can’t see the cooldown on the screen. They don’t have brains that are constantly taking the easy way out, or fingers with a built-in lag, or eyes that only look at one thing at a time. It’s not just that spreadsheets are perfect, it’s that they’re not human.

Although, I’ll say it again. The closer you get to actual combat conditions the better, so personally I’d like to see a field test go a bit farther than Ooze so you can learn more.

As already state, the human factor is pretty much same for every build in simple situations like Ooze.

I’m all for a real testing but it should be close to a real run, not close to a spreadsheet.

Are there any methods in place to calculate time-damage analysis of attack chains? Most of the stuff I’ve seen here maintain damage numbers… but correct me if I’m wrong, some (if not most) of the spreadsheet calcs shown here base themselves on average attack rotations. How about checking if the attack rotations are either front-loaded or backloaded damage, as well as if it’s either action heavy or not to maintain? For this stage, it might be best to turn off effect modifiers based on boons and conditions and focus on sole damage done of said rotation, all boosted by armor, food, slaying potions and sigils, as well as rune bonuses.

Once we can get there, we can probably model the boons, conditions, and non-permanent effects for every attack present and update the old data, because I think this can show the possible damage of a run, and how interrupting during a skill affects DPS.

This way we can check the average attack rotations, because AFAIK that’s may give us the reason why 4w/1m has greater damage because the damage is spiked in the front and easier to maintain, therefore having better efficiency, as compared to high-damage potential rotations but have skills that have many actions required, therefore increasing the effect of network-based delay on the rotations.

This is from a noob/amateur’s perspective though, so take it with a grain of salt. This kind of testing requires extensive knowledge of skill data and how they are implemented in game, as well as a lot of number crunching, since not only do you have to take into account damage and time, but also other internal things in the calculation such as boon duration of the skill, skill timings and cooldowns down to the 0.1th of a second.

P.S: With a little modification, this can be extended to PvP, simply because we need to also include animation timings of the skill, which may or may not be off by a couple of fractions of seconds.

No we really can’t use spreadsheets to model PvP, it’s simply impossible!

But you are right, paying utmost attention to the rotation calculation is crucial.

Because all dungeons involve more running than fighting. E.g taking a Mesmer gets you a kitten fast Lupi kill, but overall having one slows down the run by being a slow piece of kitten.
To answer the Thief question from the start I do believe it is CnD + Backstab (gotta love it when CnD hits harder than a face stab when you can’t get behind the target) until 50% health at which point Heartseeker damage catches up and goes through the same amount of initiative, so you’d just double Heartseeker instead of CnD whilst maintaining +6 initiative IIRC. And the glorious Assassin’s Signet + Haste + Heartseeker spam never gets old

Stands to reason if you contend there’s no difference between spreadsheet and practice in any fight that downplays movement, you’d expect those sorts of fights to pan out exactly the same down to the mili-second.

Do you actually get results like that? I can’t say I ever have.
Trivializing the movement component of a fight is practically a Ranger’s M.O., so whether I like it or not, anytime I wanna’ mess around with a build I experience that sort of a fight. Last time I beat up the Reddit Ram to check out a new build idea that had a very particular order of skills, I got the following;

3:15, 3:29, 3:16, 3:15, 3:12, 3:16, 3:07, 3:33, 3:16, 3:05*

And that’s a pretty normal thing for me to see.
Occasional spurts of bad, rare spurts of good that are more reflective of the spreadsheet but never really quite there, and then some kind of median. I haven’t been able to manage perfect clockwork millisecond similarity across all instances, and I also haven’t seen practice ever hit the napkin math even on the low-end results.

/*(Semi-reluctant to post these numbers. This is just my playing around with a build, I don’t want to see people taking it as some overall indication of Ranger’s DPS potential.)

Honestly, if you want a reasonable DPS check just use the risen abomination in arah. You can avoid the auto attacks with movement, so classes that rely on full endurance traits aren’t disadvantaged too badly. You’ll have to evade or simply aegis the jump, but thats not a big deal. The super saiyan form I guess will cause some consistency issues, but otherwise it’s pretty much the perfect fight to benchmark DPS on.

Easy boss, fast to get to, doesn’t disadvantage classes that rely on flanking/endurance traits.

Vox Hollow.2736:

Stands to reason if you contend there’s no difference between spreadsheet and practice in any fight that downplays movement, you’d expect those sorts of fights to pan out exactly the same down to the mili-second.

Do you actually get results like that? I can’t say I ever have.
Trivializing the movement component of a fight is practically a Ranger’s M.O., so whether I like it or not, anytime I wanna’ mess around with a build I experience that sort of a fight. Last time I beat up the Reddit Ram to check out a new build idea that had a very particular order of skills, I got the following;

3:15, 3:29, 3:16, 3:15, 3:12, 3:16, 3:07, 3:33, 3:16, 3:05*

And that’s a pretty normal thing for me to see.
Occasional spurts of bad, rare spurts of good that are more reflective of the spreadsheet but never really quite there, and then some kind of median. I haven’t been able to manage perfect clockwork millisecond similarity across all instances, and I also haven’t seen practice ever hit the napkin math even on the low-end results.

/*(Semi-reluctant to post these numbers. This is just my playing around with a build, I don’t want to see people taking it as some overall indication of Ranger’s DPS potential.)

And with build with twice as much DPS you would clear it about twice as fast. Any mistakes you made in one build would also appear in the other because there aren’t really any skill shots.

I’ll admit, I don’t quite follow?

Wethospu.6437:

And with build with twice as much DPS you would clear it about twice as fast. Any mistakes you made in one build would also appear in the other because there aren’t really any skill shots.

But the force that skews maths ingame …?
Could you be a non believer?
Let the force flow through you

It’s not mysticism versus math. :p
It’s math that’s taking more things into account versus math that’s taking less things into account. Am I really doing such a bad job explaining this?

Nonono don’t worry I don’t really agree but what you say makes complete sense

I just like to say stupid jedi stuff

Ah, That’s alright then.

best dps test i know of would be the abomination in Arah Path 2 when using the safespot

Vox Hollow.2736:

I’ll admit, I don’t quite follow?

Let’s say you make a build with X dps. You test it on static target and get 0.9 * X dps because of ping/lag/whatever. Then you try build with Y dps. Now if assumed that you do same mistakes/lag/whatever you only get 0.9 * Y dps.

Let’s say Y = 2 X which means that second build does twice as much damage than the first one (Y / X = 2 * X / X = 2).
Now let’s compare “real” results. (0.9 * Y) / (0.9 * X) = Y / X = 2 * X / X = 2.

So with static dps test we get exactly same result as we would have got from the spreadsheet. It obviously doesn’t go like this if one build is much harder to play than the other but most rotations are extremely simple to use.

I’m interested but I’m not sure what the motivation is to do the field test. I think our understanding of how damage is dealt and the times/durations are understood well enough to not need a field test to verify that understanding. Is there something that the originators feel is missing from that knowledge or is this a purely academic exercise to see if it can be done?

I think the variance from the time spiking up and down is the lag/mistakes/human error stuff. The fact the median is off from the spreadsheet seems to me more of a ‘realistically possible’.

You can totally predict the variance. It’s kinda’ samey, and about what you’d expect.
But, the median shift from paper to practice is a bit harder.

It just seems to come down to how badly I didn’t consider the realities of what doing something actually entails when I first made the build. It’s not like there’s a fixed penalty for effing that up. How ‘off’ you are just depends on how much you banked on a person doing something a person can’t do for a boss-length period of time, or relied on some assumption about the game that just isn’t there, or didn’t factor in one of the endless endless endless minutia that ended up being unexpectedly important.

Wanna’ just autoattack and mash a couple utilities as soon as they come off cooldown? That’s not so bad. Wanna’ have your build absolutely revolve around a 25 second cooldown attached to the weaponset you can’t see? Could be a bad idea. Wanna’ make a build that relies on the constant interruption of a back-loaded autoattack whose last hit isn’t visually or audibly distinct from the other two? Definitely a bad idea.

The only real way I know of to get some idea of what the realistic limits are is by putting builds into practice and smacking into them face-first.

/edit: terminology was a bit confusing, and probably likely to give the wrong idea. Changed for clarity.

The reason for it is because using maths, certain compositions yield better theoretical dps. But in reality stacking mostly warriors gives faster kill times. It would be good to find out why (we kind of know already that its because of easy access to buffs and vuln). But doing an actual test with different classes allows us to compare exactly how much the buffs fall off and how much dps the classes can provide compared to each other in a real environment. Also builds like my necro build are a lot harder to calculate dps in a rotation because of the amount of skills and variables that increase dps.

FGJ op

There was a thread a few months ago where people were speed killing the champ quaggan in FG, if they have a reasonably fixed HP, wouldn’t that type of set up allow for a reasonable assessment of individual contribution with relative times?

Just got a pretty good idea for a mob you could use, although it might not be as great as I think myself.

Sotzz the Scallywag? He always spawns at the same spot, all you need to do is punt him out of his shield and you’ve got a pretty weak enemy (especially if you don’t have many people) that doesn’t move or do much.

Downsides:

• need to get him, although for all the guild bounties I’ve done, Sotzz has been pretty common. If you pop t3 bounties, you’ll get him quite easily.

• open world, PUGs might screw up your test

• only a limited time outside of his shield means you need to time your CC and punts really well

Bright.9160:

Just got a pretty good idea for a mob you could use, although it might not be as great as I think myself.

Sotzz the Scallywag? He always spawns at the same spot, all you need to do is punt him out of his shield and you’ve got a pretty weak enemy (especially if you don’t have many people) that doesn’t move or do much.

Downsides:

• need to get him, although for all the guild bounties I’ve done, Sotzz has been pretty common. If you pop t3 bounties, you’ll get him quite easily.

• open world, PUGs might screw up your test

• only a limited time outside of his shield means you need to time your CC and punts really well

Doesn’t he have defiant? That would be a PITA for one person to pull off the defiant (~3) and yoink him out of his ring for a DPS test, and may lead to non-optimized CC skills being slotted. Also, this would be easier for some professions over others, in some cases, an optimized thief, for instance, can spam pistol whip to pull of stacks, but would need to slot scorpion wire to pull.

Wait… is this sarcasm? I am truly curious about this test, so I will be discussing this in earnest from now on.

lunyboy.8672:

Bright.9160:

Just got a pretty good idea for a mob you could use, although it might not be as great as I think myself.

Sotzz the Scallywag? He always spawns at the same spot, all you need to do is punt him out of his shield and you’ve got a pretty weak enemy (especially if you don’t have many people) that doesn’t move or do much.

Downsides:

• need to get him, although for all the guild bounties I’ve done, Sotzz has been pretty common. If you pop t3 bounties, you’ll get him quite easily.

• open world, PUGs might screw up your test

• only a limited time outside of his shield means you need to time your CC and punts really well

Doesn’t he have defiant? That would be a PITA for one person to pull off the defiant (~3) and yoink him out of his ring for a DPS test, and may lead to non-optimized CC skills being slotted. Also, this would be easier for some professions over others, in some cases, an optimized thief, for instance, can spam pistol whip to pull of stacks, but would need to slot scorpion wire to pull.

Wait… is this sarcasm? I am truly curious about this test, so I will be discussing this in earnest from now on.

Nope, it wasn’t sarcasm, was just throwing out ideas.

Here’s an engineer vulnerability build. Not really able to do the math for vuln at work, at least not right now.

http://gw2skills.net/editor/?fcEQJAqel0piYX3y2F1LJx4DdGkCAbXRFy/XG0PtA-jACBYNBk0AkSAINqIasFhFRjVFBTriIq2JoIa1CBsYBA-e

Very low power, despite exotic armor, will provide 3 stacks of might and fury about every 10 seconds, as well as maintaining around 20 stacks of vulnerability with the +85% condition duration.

I did use giver’s exotic weapons, though, since it doesn’t matter what weapons you use on Engineer, the kit damage will always be BIS (exotic) at that level. You could opt for a level ~60 engineer, though, that would have access to the traits, runes and sigils, but not the exo gear and wouldn’t have the BiS issue with kits.

The rotation would ideally be GK 1, GK3, TB-Analyze, GK 1…

With alternatives being switching to elixir gun and using fumigate, or slotting net turret to proc vuln on immobilize.

There is some necessary math to determine which rotation would be ideal while doing the absolute least damage.

This does rely on crit procs for vulnerability in both a sigil and a trait, and that could be changed if there is even more damage reduction needed.

Edit: wording.

that engi build sucks

hybrid.5027:

that engi build sucks

For damage, yes.

For vulnerability? It’s amazing.

The OP was asking if an engineer could keep up 25 stacks of vulnerability. I think this build can do that, with a bit more theory-crafting.

Finally got out of the SHU, wahoo!

Wethospu.6437:

Let’s say you make a build with X dps. You test it on static target and get 0.9 * X dps because of ping/lag/whatever. Then you try build with Y dps. Now if assumed that you do same mistakes/lag/whatever you only get 0.9 * Y dps.

Maybe I missed it, but where are these tests? That’s exactly what I’m advocating doing! I get your point that it would be redundant if it matched up exactly with the spreadsheet math, but as far as I know no “big-picture” correlative tests whatsoever have ever been done to confirm the spreadsheet results of any build comparisons.

And if you think that it’s a pointless waste of time because the spreadsheets are perfect, all I’m asking you to do is prove it.

lunyboy.8672:

There was a thread a few months ago where people were speed killing the champ quaggan in FG, if they have a reasonably fixed HP, wouldn’t that type of set up allow for a reasonable assessment of individual contribution with relative times?

I think the Quaggan is probably too aggressive to take out the high variables of individual skill. My times versus Wethospu’s time would most likely be drastically different even with the same build. Not to mention the issue of people getting in your way, since it’s part of the idiotic supa-casual champ train.

I’d use bjarrl. He’s just a big punching bag when you fruit him and you could flank him if needed or whatever for the test and you can’t do that to the ooze. It’s late in the path but I think it fits your criteria the best

Broadicea.8294:

Wethospu.6437:

Let’s say you make a build with X dps. You test it on static target and get 0.9 * X dps because of ping/lag/whatever. Then you try build with Y dps. Now if assumed that you do same mistakes/lag/whatever you only get 0.9 * Y dps.

Maybe I missed it, but where are these tests? That’s exactly what I’m advocating doing! I get your point that it would be redundant if it matched up exactly with the spreadsheet math, but as far as I know no “big-picture” correlative tests whatsoever have ever been done to confirm the spreadsheet results of any build comparisons.

And if you think that it’s a pointless waste of time because the spreadsheets are perfect, all I’m asking you to do is prove it.

What do you think might go wrong?

Damage formula is incorrect? I’m quite sure that lots of people have been testing that.

Animation durations are wrong? This is a possibility but testing on Mists would be sufficient for this.

Making sure animation durations are right is the tricky part.
For the elem we have DEKeys working on specific skill’s animations and Sabull working on full rotations, the end result is a very accurate picture.
I don’t know what could go wrong in practice … if the situation assumptions of the spreadsheets are correct and match that of reality then combat should develop as predicted.
This is unless the force truly intervenes in-game of course

Wethospu.6437:

What do you think might go wrong?

Damage formula is incorrect? I’m quite sure that lots of people have been testing that.

Animation durations are wrong? This is a possibility but testing on Mists would be sufficient for this.

Nothing, I just value empirical results over abstract hypotheses.

Put it this way: if this game had a Master of Damage, would you use it to verify? Or just trust the spreadsheets?

What you describe with this diagram is what he have done separately for every single element in our spreadsheets.
We have every formula, we have tested the relationship between formulas, so unless there is a mystical force in-game, or we are missing information, then there is no way our results will be incorrect.
Now you may say that we could be missing information, but honestly theorycrafting for this game is really straightforward: there are only 2 or 3 formulas involved … so I think we got it right
But if you insist, then let’s do a “half” test: let’s design a simple rotation, a build with PvP gear, and try this in the mysts on the unbreakable golem.