Rawr Optimizer/Graph Differences?

[Walookani]

Hey Shamans,

I'm still relatively new to the graphing, simulating, and heavy testing of my DPS, but im getting there. I had a question regarding Rawr.

When I view my Stats Graph, (currently) it says that Attack Power weighs more than what i was stacking before, which was Hit (it was higher in both the graph, and it came up in the Optimizer).

Therein lies my question: why is it that the Optimizer relies so heavily on Hit when "Optimizing DPS," and the graph clearly shows that Attack Power would yield stronger/more DPS?

This may be something silly I've done or overlooked or and am just not understanding, so i was hoping I could see some nice answers.

(PS--I do test out my stuff in EnhSim that has been exported from Rawr)

[Deleted]

At the heart of Rawr is a calculation engine for each spec. Rawr takes a bunch of stats, including buffs, procs, etc and throws them at the calculation engine and gets back a figure. So the calculation engine basically crunches all your stats into a single dps figure.

The optimiser uses this to simply take various combos of gear, it generates the stats for that gear and says to the calculation engine give me a figure. The optimiser then tries other combos and does the same. Each time its looking for a higher figure as a result. It settles on the gear that gives it the highest figure. Therefore the optimiser doesn't give a fig about what stats the gear has all it is doing is getting a figure from the calculation engine.

Now the graph does the same it takes your current stats adds some AP, or hit, or haste or whatever line its drawing, and throws that new stats with the added AP, hit, crit whatever and gets back a new figure. It basically does this for +/- 100 of each stat and draws a line representing the figures it got back. As you can see once again it doesn't give a fig about what stats it has its only throwing a bunch of stats at the calculation engine and getting back a figure as a result.

Therefore both methods are using EXACTLY the same calculation engine and thus should be giving the same result. Can you give an example of what is different please, it might then shed light on why you are seeing what you are seeing. BTW what version are you using?

[Walookani]

I am using Rawr version 2.3.4, what i believe is the latest.

Let me say this, first: Sometimes, if i Optimize my gear, get a result, and then Optimize that RESULT, i get different Optimization, usually consisting of different stats, but the DPS rate is the same. An 'example' might be one optimization saying a gem socket should be +20 Hit, and the the one after having that same socket with +20 AP/+10 Hit.

I'm unsure of what exactly what i can give you when ask for an example of what is different (between the graph vs Optimizer) All i can say that when i optimize, as of late, the Optimizer shows that veritable stacking Hit will yield the greatest DPS, as every socket and every enchant is practically Hit. However, when one looks at the Graphed Data, the 'line' for Hit is clearly in the middle of "DPS increase quadrant," whereas Attack Power is clearly above Hit, Haste, and Crit.

I was just wondering why, if the Graph measured AP as such a solidly better choice for a DPS increase, the Optimizer chooses Hit, or (vice versa) why if the Optimizer chooses Hit for every socket and enchant, the Graph measures AP as best.

[Deleted]

If you could provide before and after XML files and post them as an issue at http://rawr.codeplex.com/WorkItem/List.aspx please.

I can then see the pre-optimised values and post-optimised values to assist working out what has been done.

The reason you don't get the same answer every time when re-optimising is that the optimiser doesn't run through every last possible combination of gear. eg: lets just take a simple example where you have only marked with a green diamond 1 item in each slot and 1 enchant per slot - so far nothing to optimise. Now lets say you go with the default gemming templates and have say 10 gem slots. It is going to try around 12 different gemmings in each of those 10 slots. So that's a possible 12^10 different combinations which is 61,917,364,224 ie: almost 62 billion possible combinations to test. If it did test them all it would never finish in your lifetime. So instead it uses very clever maths to try to hone in on the solution. This is called a hill climber solution where if you imagine all the combinations plotted on a XY grid with dps as the height at that point you get a terrain of hills and valleys.

The hill climber starts at a random point and trys to move to a nearby solution to find if its "climbing a hill". It climbs that hill trying to find the top. Periodically it will randomly jump to another part of the map to see if it can find a bigger hill or even a huge mountain thats a big dps increase. How effective it is in finding the highest hill depends on how flat the terrain is and if there are any really big peaks to find. If all the options give pretty similar results it can go round in circles finding little hills all of similar heights, and it just picks one as the answer.

Hope this helps explain it a bit. However do post an issue about the hit I had nailed that in the past so its odd it resurfaces.

[Walookani]

I'll try to do that for you.

Now that i think of it, i remember posting before and you, Levva, were the one who showed me that you can look at the data in a graph.

I remember i was getting Hit and Haste for my optimization, and i had "hills and valleys" in my Graph between Hit and Haste, so Rawr was at least correctly correlating between two sets of Data.

Now, however, the AP/Hit thing, it seems its buggin' out!

[Deleted]

I changed the calculation methodology to avoid the dramatic swings that haste in particular was giving so the graphs should be a lot smoother now.

[detramental]

this may be out in left field, but do the dips and valleys in the haste graph represent DR after you exceed 1 second in cast time?

please correct me if i'm wrong. i was also a bit fuzzy on viewing the graphs and their meanings.

gradual up-climbing graphs are pretty self explanatory, but when you get dips it makes you wonder what causes the decrease.

[Deleted]

this may be out in left field, but do the dips and valleys in the haste graph represent DR after you exceed 1 second in cast time?

please correct me if i'm wrong. i was also a bit fuzzy on viewing the graphs and their meanings.

gradual up-climbing graphs are pretty self explanatory, but when you get dips it makes you wonder what causes the decrease.

No they represent the fact that you can get X attacks of one ability say Earth Shock off in a 7 min combat period with a 1.38 (random number) GCD. Now it might be that you can add haste and it will not change the number of Earth Shocks you can get until suddenly 1 extra haste means you can squeeze in an extra Earth Shock in the 7 mins and suddenly you have X+1 ES in the time instead of X. This gives a spike in dps.

Other effects of haste can give dips where by hasting just that tiny bit more that a MW stack becomes available a few hundreds of a second earlier and so pushes other abilities back by a GCD. So the net effect is actually you do LESS attacks of that ability in the 7 mins combat. This is why the sim was giving haste as very spiky. Using the old "exact" model methodology. The newer "smoothed" methodology works out an average number of attacks per combat cycle so you never get exactly X or X+1 but fractional increments. This has smoothed out the graphs and so ALL graphs in Rawr should now be relatively smooth. If they aren't there is something seriously wrong and I need to know about it.