Register

Stat weights

Face-rippin fun.

Moderator: Forum Administrators

Honored
Posts: 82
Joined: Mon Jan 03, 2011 7:59 am

Re: Stat weights

Postby Arctagon » Sun Jan 27, 2013 10:48 am

raffy wrote:X >= X as in "Assuming no procs, your new Rake is (X) always as good (>=) as your current Rake (X)" which would imply Rake spamming.

Oh, right. X >= X in itself doesn't make any mathematical sense, so it made me confused there for a second.

No, this does make sense: the lower the ratio, the more likely your going to clip Rake. You energy regen is basically a constant for a given fight. So if your Raking more, your overall Rake % must increase.

Yeah, but when the ratio is lower, the DPS gain is higher, which means that you get more DPS out of Rake, which shouldn't result in Rake constituting less of our DPS. As you said yourself, the more we Rake, the more the overall Rake percentage increases.

Equally strong (Ratio >= 1.0) is the left side of the discontinuity; Clipping anything better (Ratio > 1) is the right side.

What, when the ratio is greater than or equal to 1.00, it's either exactly on 1.00 or to the right side of the discontinuity, not the left? The blue part of the graph, in other words. According to the graph, one of the highest DPS gains is at the ratio 1.00 (the highest being at approximately 1.02), so you could essentially spam Rake seeing that you don't get any procs, and it would be a DPS gain.

That was the theory before, when the ratio was 1.12 -- that's really difficult to do in your head. Ratio > 1 is much easier to approximate: anytime you get a proc, apply Rake. I'll sim this one when I get a chance.

Is it really that much easier? If you have a TFed Rake up, and then your trinket procs, it's not always a given if applying a new Rake is going to make it stronger or weaker.

We should move this discussion to the other thread.

We should. I don't know which thread you're referring to, though, so I can't take the initiative.


aggixx wrote:Yeah, those numbers are just stats from gear. Armory also accounts for base stats and I think it also accounts for Leather Specialization but I'm not 100% sure.

There's a "raid-buffed" stats column under stats in the HTML report, that's what you want to look at for comparison..

The base stats I can probably understand, but the secondary stats? As far as I'm aware, we get the secondary stats from the gear only, so it should be consistent with Armoury. The reason I brought this up is because it's useful to change the stats before simulation if one wants to simulate with reforges without going in-game and wasting gold on a reforge, just so you can use it for a simulation. That's going to be very expensive in the long run, let alone time-consuming. You could argue that one can simply import profiles from other sources, but with CharDev being completely useless and Rawr not being able to reforge gear (? I'm honestly not completely sure, as I no longer use Rawr. Even if it is possible, it's kind of retarded that I need an entire program installed to my computer just so I can simulate reforges (especially when the program is meant for so much more -- it's massively overkill), on top of it being cumbersome and time-consuming), I don't have any other alternatives that I'm aware of.

Exalted
User avatar
Posts: 894
Joined: Tue Oct 23, 2012 7:15 am

Re: Stat weights

Postby raffy » Sun Jan 27, 2013 6:17 pm

Are you intentionally taking the opposite position of every single point made in this thread? Many of us spend our free time commenting here because we're interested in improving our class and enjoy theorycrafting and figuring stuff out.

Possibly my plot is not clear enough. I only plotted a few things. I updated the plot with some more data. The plot now consists of 162 simulations of 250k (40M+ sims). I tried to resolve the region between 1 and 1.1 a bit more.
http://raffy.antistupid.com/wow/catsim/ ... plier5.png

You should not extract exact values from the plot, there is no value in saying, that a Ratio of 1.01441 is better than a Ratio of 1.01641, this is overfitting and the result of noise. I fit a simple polynomial to the highest DPS region ( http://raffy.antistupid.com/wow/catsim/ ... plier6.png ) but I don't see any reason it should have to follow this shape other than it does.

Yeah, but when the ratio is lower, the DPS gain is higher, which means that you get more DPS out of Rake, which shouldn't result in Rake constituting less of our DPS. As you said yourself, the more we Rake, the more the overall Rake percentage increases.

Except that every time you Rake, you aren't doing something else (finite amount of energy in a given fight), so while your Overall Rake damage increases, your Overall Total damage decreases. That is the purpose of this analysis: is there a time, if at all, that we should be clipping an existing Rake? And the results show: you get more DPS out of Rake as long as you clip with a Ratio > 1. The discontinuity is evidence of Blizzard fixing the Rake/Shred balance (at least for the current level of Mastery), since before that changed, Rake spamming was shown (at some reasonable amount of Mastery) to beat Shred.

What, when the ratio is greater than or equal to 1.00, it's either exactly on 1.00 or to the right side of the discontinuity, not the left? The blue part of the graph, in other words. According to the graph, one of the highest DPS gains is at the ratio 1.00 (the highest being at approximately 1.02), so you could essentially spam Rake seeing that you don't get any procs, and it would be a DPS gain.

According to the graph, there are 2 points at a Ratio of 1.00 and I've been trying to point out: one is a special case, and it corresponds to Ratio > 1, and the other is the expected Ratio >= 1. I changed the color of the regions to illustrate the difference in meaning (Rake clip vs Rake spam).

Honored
User avatar
Posts: 67
Joined: Sat Dec 08, 2012 3:35 pm

Re: Stat weights

Postby Vami » Wed Jan 30, 2013 3:20 am

@raffy:

I already mentioned this in another thread but I'm not sure if you saw it there or not. Anyhow, what came to my mind while looking these graphs was that refreshing Rake at 1.00 should allow you to have more ticks with procs on than leaving it only to 1.12< or so. This could be because you'd also refresh the bleed just before the procs fell off, instead of refreshing it only when you'd get a lot more powerful Rake on. Waiting would probably result in a longer time with a less powerful Rake whereas replacing the 1.00 rated just before the procs fall would let you keep a powerful Rake up longer, since it was refreshed just before the buffs fall off. If you wouldn't refresh it with procs up at the last second (at 1.00 ratio), it'd fall off and you'd need to apply a new one faster, probably without any procs up at all.

This theory is, of course, no good if you already have some kind of a apply-bleeds-again-if-procs-are-about-to-fall-off line in your sim BUT if you don't have one, I'd really like to see how the Rake dps would act with that one. I mean that you could have a separate line for applying Rake again with procs about 1-4 secs before the mentioned procs fell off and then a separate one for actually overwriting Rake because of getting some procs that'd make the Rake (for example) over 1.12 times better than the pre-existing one.

Exalted
User avatar
Posts: 894
Joined: Tue Oct 23, 2012 7:15 am

Re: Stat weights

Postby raffy » Wed Jan 30, 2013 7:48 am

I have support for lookahead Rake and Rip, but I am not using it. I am simulating aggixx's actionlist as-is with my own simc-ish interpreter that I hacked into Catus.

In the simc actionlist, we're currently using the following statement to express the ratio of Rake tick damage:
Code: Select all
tick_multiplier%dot.rake.multiplier>___

Where "tick_multiplier" is really just "Tick Damage of Rake if I apply now" and "dot.rake.multiplier" is "Tick Damage of Existing Rake"

Lookahead support can be exposed to the script by thinking of tick_multiplier as a function of time (in the future), where the current tick_multiplier would correspond to 0-sec lookahead.

For example, the following could represent the ratio of a new Rake to a Rake performed 3-seconds from now:
Code: Select all
tick_multiplier[0]%tick_multiplier[3]>___


I agree with your statement: there should exist another /rake action that utilizes lookahead to clip Rake based on proc/modifier expiration. The best example being right on the pull: you proc/use like 6+ things, they all fade in 12sec or so, so there is a catastrophic tick_multiplier decay in a very short window. At 45000 buffed AP, TF+DoC+SR+Bottle+Spring+DancSteel+Pot is a 2.82x Rake modifier, and 25-30 into the fight second mark, we basically have 0 modifiers active. Missing that Rake clip before everything fades is brutal.

If you'd like to suggest a few permutations of /rake conditions using the tick_multiplier[t] notation, I'd gladly simulate them.

Possibly, you need more data exposed, like "dot.rake.remainingDamage" (or "dot.rake.remainingTicks") and "totalDamage[t]" so you could figure out some kind of cost/benefit based on remaining damage vs new damage.

I encountered two strange things when I developed lookahead. First, I made it assume that if Savage Roar would fade within the requested horizon, that it will be rebuffed, so it is not included in the lookahead decay. This prevents unnecessary neutral rake clipping ("oh shit, the next rake is 30% weaker, quick Rake!, SR falls off, SR reapplied, doh!"). Second, since DoC has charges, any lookahead statement probably needs different logic depending if 1 or 2 stacks are available. Since if you have 1 stack of DoC, and you don't bother to clip the Rake, and instead execute a filler action (Shred), you now have an unexpected damage dropoff. Possibly, since DoC lasts 30 seconds (which is far beyond the typical lookahead of 1-6 seconds), that instead of using the expiration, it should pretend each stack of DoC is worth a 1-2 seconds of duration.

Previous

Return to Kitty DPS

Who is online

Users browsing this forum: Bing [Bot], Google Feedfetcher and 7 guests