Elitist Jerks On the Value of Haste

02/06/10, 8:22 AM   #16
Aiel
Von Kaiser

Dwarf Priest

Destromath (EU)
 Originally Posted by kamaia It is not true that the DA proc is "less likely" to overheal. Rather, the DA proc cannot overheal. The question is really whether the shield will be consumed before it expires. As the original post points out, there are two principal roles for the discipline priest in the raid: tank healing with a PW:S/PoM/Pen/FH rotation or shield-spamming the raid during heavy, predictable raid damage (e.g., Prof. Putricide phase 3, Blood-Queen). I will consider the first case since neither crit nor haste really helps for shield-spam encounters.
The DA proc can overheal, when you reach the 10k absorb cap and in some cases DA can even expire due to tank swaps or phase transitions.
Typically you need 3 FH crits to reach a 10k+ DA shield, with 40-50% crit this occures around 6.4-12.5% of the time and takes around 2.3s without damage on the tank. It is not much but it happens, especially under the effects of power infusion or heroism.

 02/06/10, 7:33 PM #17 Baptistin Glass Joe   Baptistine Human Priest   Dunemaul Hey all. I uploaded my log from ICC10 entirely as disc (spare valithria) for you all to use... for science! http://public (dawt) me (dawt) com/jomanscool2 (for some reason it was adding auto links to my website, hence the added code tags). I have all my raids logged on my hard drive for the last three months if a different log would be better suited for this testing. Also, if for some reason the rar-ed version of the log is not good enough, I can upload the 130MB logs in their whole glory without a problem.
02/08/10, 2:29 PM   #19
kamaia
Glass Joe

Troll Priest

Jaedenar
 Originally Posted by Baptistin Hey all. I uploaded my log from ICC10 entirely as disc (spare valithria) for you all to use... for science!
Thanks, I'm sure this won't be the last time I use it.

In the interest of good scientific practices the code I used is below:

 #!/usr/bin/perl use strict; my $char = '"Baptistine"'; # Character to analyze (use quotes) my ($heal, $overheal,$crit, $overcrit) = (0, 0, 0, 0); while (<STDIN>) { next unless /SPELL_HEAL/; my @event = split /,/; next unless$event[2] eq $char; if ($event[13] =~ /nil/) {     $heal +=$event[10];     $overheal +=$event[11];     } else {     $crit +=$event[10];     $overcrit +=$event[11];     } } print "Regular: $overheal /$heal (" . 100 * $overheal /$heal . "\%)\n"; print "Crit: $overcrit /$crit (" . 100 * $overcrit /$crit . "\%)\n";  
So with all of the usual disclaimers about sample size, etc., from this parse the values I get are 32.8% overheal on regular, 45.5% overheal on crits. In playing with these numbers, I realized that my earlier formula for the effectiveness of crit has problems. To try to make sense of these numbers I'm going to use the formula below for the percent increase in effective healing from 1% crit. C is current crit percentage, a_r is the regular overheal rate, a_c is the critical overheal rate, m is a binary variable indicating the presence of the meta, and k is a regularization constant to account for haste, spellpower and basically anything else other than crit percentage.

$\frac{k(0.99-c)(1-a_r) + k(0.01+c)(1.5(1-a_c)+0.45)}{k(1-c)(1-a_r) + k(c)(1.5(1-a_c)+0.45)}-1$

Using the values supplied I get 0.65%

Assuming these values are also reasonable assumptions for holy (which is a big assumption):

$\frac{k(0.99-c)(1-a_r) + k(0.01+c)(1.5(1-a_c))}{k(1-c)(1-a_r) + k(c)(1.5(1-a_c))}-1$

yields 0.20%. The file I parsed also contained a holy priest for a few fights, the results there were c=.23, a_r=.42, a_c=.53 and this also yields very close to 0.2%.

There are a few things that are important to note in these calculations. First these are not pseudo-power weightings or anything like that. A value of 0.65% means adding 1% crit should increase effective healing by 0.65%. 1% increase in spellpower (~33 for the character being parsed) should increase effective healing by a little more than 0.48% (adjusted for the frequency of overheals since SP doesn't increase the amount healed on an overheal). It's "a little more" because even on an overheal, DA is increased with spellpower.

I would like to do the spellpower calculation more carefully and work in the effect of the [Revitalizing Skyflare Diamond], but I'm going to hold off in case someone finds a mistake in all of this. This suggests that 1 point of crit rating is worth about 1 point of spellpower for disc and 0.3 spellpower for holy.

I think this form of analysis has a lot of potential though. Examining parses and trying to guess what throughput would have been if stats were different could be a very powerful technique -- similar to simcraft but for healing.

---
Edit: I found a mistake in the spellpower calculation. It should be correct now, the numbers make a lot more sense.

Last edited by kamaia : 02/08/10 at 5:23 PM.

 02/08/10, 4:23 PM #20 Nicene Glass Joe   Nicene Night Elf Priest   Eitrigg Before I do some crunching on my end, can I get some confirmation on relevant characteristics for DA? 1. DA total is capped at 10000 2. DA lasts 12s and that is refreshed each time the effect occurs 3. DA is absorbed "first", that is before Sacred Shield, PWS, AMS, etc? 4. All critical heals contribute to DA, regardless of overheal (I just tested this in game so it should be the case). 5. The combat log does not differentiate between absorbs. Sorry if this is remedial, but I'm trying to write a quick parser for DA caps and I want to make sure my assumptions are well founded.
02/08/10, 4:51 PM   #21
kamaia
Glass Joe

Troll Priest

Jaedenar
 Originally Posted by Nicene 1. DA total is capped at 10000 2. DA lasts 12s and that is refreshed each time the effect occurs 3. DA is absorbed "first", that is before Sacred Shield, PWS, AMS, etc? 4. All critical heals contribute to DA, regardless of overheal (I just tested this in game so it should be the case). 5. The combat log does not differentiate between absorbs.
I can confirm 2, 4, and 5. 1 and 3 should be determinable from the combat log and some creative parsing. I remember seeing a blue post on order saying that it's either a last-in-first-out (LIFO) or first-in-first-out (FIFO), but I can't find it now.

02/08/10, 5:23 PM   #22
Squeakster
Piston Honda

Dwarf Priest

Eitrigg
 Originally Posted by Aiel The DA proc can overheal, when you reach the 10k absorb cap and in some cases DA can even expire due to tank swaps or phase transitions. Typically you need 3 FH crits to reach a 10k+ DA shield, with 40-50% crit this occures around 6.4-12.5% of the time and takes around 2.3s without damage on the tank. It is not much but it happens, especially under the effects of power infusion or heroism.
This is a good point about the 10k cap on DA, although in order for this to come into play while chain casting FH we would need to consider the fourth consecutive crit, not the third, because the fourth DA is the one that would be wasted. Assuming 1.2 second Flash Heals and 40% crit, the 10k cap should only come into play about 2.6% of the time and only if the tank takes no damage over the 3.6 seconds of those Flash Heals. Certainly something for us to keep in mind, but it doesn't sound like it would throw off the value of crit very much.

 Originally Posted by Baptistin Hey all. I uploaded my log from ICC10 entirely as disc (spare valithria) for you all to use... for science!
Thank you for the data sir.

-----------

02/08/10, 6:35 PM   #23
Squeakster
Piston Honda

Dwarf Priest

Eitrigg
 Originally Posted by kamaia Thanks, I'm sure this won't be the last time I use it. In the interest of good scientific practices the code I used is below: ... So with all of the usual disclaimers about sample size, etc., from this parse the values I get are 32.8% overheal on regular, 45.5% overheal on crits. In playing with these numbers, I realized that my earlier formula for the effectiveness of crit has problems. To try to make sense of these numbers I'm going to use the formula below for the percent increase in effective healing from 1% crit. C is current crit percentage, a_r is the regular overheal rate, a_c is the critical overheal rate, m is a binary variable indicating the presence of the meta, and k is a regularization constant to account for haste, spellpower and basically anything else other than crit percentage. $\frac{k(0.99-c)(1-a_r) + k(0.01+c)(1.5(1-a_c)+0.45)}{k(1-c)(1-a_r) + k(c)(1.5(1-a_c)+0.45)}-1$ Using the values supplied I get 0.65% Assuming these values are also reasonable assumptions for holy (which is a big assumption): $\frac{k(0.99-c)(1-a_r) + k(0.01+c)(1.5(1-a_c))}{k(1-c)(1-a_r) + k(c)(1.5(1-a_c))}-1$ yields 0.20%. The file I parsed also contained a holy priest for a few fights, the results there were c=.23, a_r=.42, a_c=.53 and this also yields very close to 0.2%. There are a few things that are important to note in these calculations. First these are not pseudo-power weightings or anything like that. A value of 0.65% means adding 1% crit should increase effective healing by 0.65%. 1% increase in spellpower (~33 for the character being parsed) should increase effective healing by a little more than 0.48% (adjusted for the frequency of overheals since SP doesn't increase the amount healed on an overheal). It's "a little more" because even on an overheal, DA is increased with spellpower. I would like to do the spellpower calculation more carefully and work in the effect of the [Revitalizing Skyflare Diamond], but I'm going to hold off in case someone finds a mistake in all of this. This suggests that 1 point of crit rating is worth about 1 point of spellpower for disc and 0.3 spellpower for holy. I think this form of analysis has a lot of potential though. Examining parses and trying to guess what throughput would have been if stats were different could be a very powerful technique -- similar to simcraft but for healing. --- Edit: I found a mistake in the spellpower calculation. It should be correct now, the numbers make a lot more sense.
Thank you for doing the leg work on this, it is definitely outside my area of expertise. All of your math looks good to my untrained eye.

One question, how did you translate the 33 SP increase into a 0.48% increase in effective healing? I'm thinking this would be a complicated calculation because of the different coefficients that different spells have.

And while I agree with your conclusion that the data shows that 1 point of crit rating is worth about 1 point of spell power for disc, I just want to point out to the general readers of this thread looking for guidance that spell power and combat ratings have different item budgets, and when that difference is taken into account what you are really choosing between is 1/0.86 = 1.16 spell power versus 1 crit rating. Although that does not take into account the benefit DA gives to smoothing about damage spikes on the tank.

I really like your thinking about examining combat logs directly to determine the value of different throughput stats; I think Nicene was beating around the same bush in an earlier post as well. I think it would be even better than simcraft because the results would be tailor made for the player, not just based on a set of assumptions and the players gear. We have always attempted to calculate stat weightings based on a large set of assumptions and then throw in an asterisk, saying the real value depends on your particular play style, raid makeup, etc.

One can imagine a website or downloadable program where you could upload your combat log and have it spit out specific stat weightings for your exact performance. Or even better, it could save every log you upload (the relevant parts at least) and base the calculations on multiple raids, multiple days, etc., for an even more comprehensive analysis.

Specific results for the individual player, not just a general stat weight.

Although that sounds like more work than I am interested in doing, haha.

----------

 Originally Posted by Nicene Before I do some crunching on my end, can I get some confirmation on relevant characteristics for DA? 1. DA total is capped at 10000 2. DA lasts 12s and that is refreshed each time the effect occurs 3. DA is absorbed "first", that is before Sacred Shield, PWS, AMS, etc? 4. All critical heals contribute to DA, regardless of overheal (I just tested this in game so it should be the case). 5. The combat log does not differentiate between absorbs. Sorry if this is remedial, but I'm trying to write a quick parser for DA caps and I want to make sure my assumptions are well founded.
I concur with Kamaia that numbers 2, 4 and 5 are correct. I have not personally tested number 1, but the patch notes from 3.1 state:
 Divine Aegis: Divine Aegis effects will now stack, however the amount absorbed cannot exceed 125*level (of the target). It will also now take into account total healing including overhealing.
Which at level 80 would be 125*80 = 10,000. I assume this is still how it works, unless they stealth changed it at some point.

I'm actually not sure about which absorb effects are used up first, that would be a good experiment.

02/08/10, 9:42 PM   #24
Richelieu
Von Kaiser

Area 52
 Originally Posted by Squeakster One can imagine a website or downloadable program where you could upload your combat log and have it spit out specific stat weightings for your exact performance. Or even better, it could save every log you upload (the relevant parts at least) and base the calculations on multiple raids, multiple days, etc., for an even more comprehensive analysis.
Maybe I'm missing something, but -- isn't the problem with this approach that it will produce stat weightings biased toward the player's preconceived notion of what spell mix/rotation is optimal?

02/09/10, 12:07 AM   #25
kamaia
Glass Joe

Troll Priest

Jaedenar
 Originally Posted by Richelieu Maybe I'm missing something, but -- isn't the problem with this approach that it will produce stat weightings biased toward the player's preconceived notion of what spell mix/rotation is optimal?
Yes, if it were implemented ideally, it would give stat weightings for the conditions under which an individual heals. It would not say that he/she was casting the wrong spells -- just how to gear for the wrong spells he/she's casting.

 Originally Posted by Squeakster Thank you for doing the leg work on this, it is definitely outside my area of expertise. All of your math looks good to my untrained eye. One question, how did you translate the 33 SP increase into a 0.48% increase in effective healing? I'm thinking this would be a complicated calculation because of the different coefficients that different spells have. And while I agree with your conclusion that the data shows that 1 point of crit rating is worth about 1 point of spell power for disc, I just want to point out to the general readers of this thread looking for guidance that spell power and combat ratings have different item budgets, and when that difference is taken into account what you are really choosing between is 1/0.86 = 1.16 spell power versus 1 crit rating. Although that does not take into account the benefit DA gives to smoothing about damage spikes on the tank.
Yeah, let me just say very clearly that everything I'm posting in this thread is a work in progress. If you're just looking for a quick answer (rather than interesting questions) go with the guide and not my post above. At some point I hope to feel strongly enough about this approach that it gets incorporated into the guides, but it's not there yet. I do believe that crit is very strong for disc, but this should be vetted by more people than me before it becomes gospel.

The spellpower calculation was very back of the envelope and I'd like to do it "correctly" at some point. I thought I made that clear in the post, but re-reading it now I see I should have been more explicit about this. I made two assumptions in coming up with the 33 SP -> 0.48% ratio. The first one, which is probably not very big, is that I ignored the boundary cases where a non-overheal gets turned into an overheal. The second is I ignored the base heals of spells, so I simply said that 1% increase in spellpower would increase all non-overheals by 1%. 52% of the direct heals in the parse had some overhealing, and I looked up the disc priest from the parse on the armory and he had ~3300 sp unbuffed.

02/09/10, 1:49 AM   #26
mutagen
Don Flamenco

Dwarf Priest

Proudmoore
 Originally Posted by Squeakster One can imagine a website or downloadable program where you could upload your combat log and have it spit out specific stat weightings for your exact performance.
Before you go too far down this path you might check out shaman_hep and the accompanying EJ thread, It appears to be fairly well regarded and may even serve as a decent starting point for a priest tool.

 Originally Posted by DeeNogger My two (not-so-informed) sents.

02/09/10, 9:07 AM   #27
Elimbras
Don Flamenco

Dwarf Priest

Eitrigg (EU)
 Originally Posted by Squeakster - Holy concentration uptime : This is basicaly the same idea but for holy and regen, haste will also increase your chance to keep holy concentration up (depending on how you manage your prom/coh/filler use of course). And it scales with crit for that matter as well. So it's not a straigth mana consumption increase there.Good point again, and I will add this to my original post.
I really don't like the kind of argument saying "Haste increases the number of heals you cast, therefore your holy concentration uptime".
Technically, this is really true.
But the effect is marginal, and people shall not count on it. If you want to increase your holy concentration uptime, you gear for crit, and/or you change your spells selections (with short spells that can trigger it). You don't gem for haste.
The gain in holy concentration uptime is far from the loss of the increased number of cast.

Just for a quick numerical evaluation :
HC is an 8s buff that procs after any crit.
I'll take the best situation for haste : the charater currently has no haste, 30% crit, and can choose between 10% haste and 10% crit.
He maintains a cycle of 1 POM - 1 COH - Renew - Renew - SOL-FH, that takes roughly 8s (including lag).
In 8s, 2 spells can triggers HC.
The current downtime is 0.7^2 = 0.49. Uptime is 51%.

With 10% more crit, downtime is 0.6^2 = 0.36. Uptime is 64%. The gain is 13% uptime.
With 10% haste, the character can know cast 2.2 HC-friendly spells per 8s. Downtime is 0.7^2.2=0.456. Uptime is 54.4%. The gain is 3.4% uptime for 10% haste, in a really favorable scenario.

02/10/10, 3:22 AM   #28
Baptistin
Glass Joe

Human Priest

<DND>
Dunemaul
 Originally Posted by Elimbras I'll take the best situation for haste : the charater currently has no haste, 30% crit, and can choose between 10% haste and 10% crit..
Only error I see in your statement is the 10% crit vs 10% haste. Haste and crit have different rating numbers for 1% gain. Therefore we should be comparing say 200 crit rating to 200 haste rating, as items are weighted based upon rating numbers not percentages.

 Originally Posted by kamaia I do believe that crit is very strong for disc, but this should be vetted by more people than me before it becomes gospel.
I'm following this discussion mostly to try and verify my belief in this as well. Think I might have to finally resocket crit for int for the LK.... tearing my mana pools apart.

 02/10/10, 8:06 AM #29 enqi Glass Joe     Criticized Worgen Warlock   Frostmane (EU) It feels like the thread has drifted off abit. Remember, You "can only" heal as much as there is need for which is why I'd consider discussions as this one to be pretty useless. Healing has too many factors involved to say "this is the best" for example, what healing setup, what encounter, how good is your group (bad tank/people tanking fires). Discussions as this one is really only viable for one thing which is heavy aoe encounters which of we only have two bosses (in ICC) of excluding a few phases such as "phase 1" of Festergut etc. As I've told every priest who've asked me for tips so far including guildies asking for tips, the best way is to try and balance things, Holy wise a decent balance of Int, Spi and Haste (in my opinion) is the stats to try and balance whilst crit and spellpower fall abit behind in the current content. When you reach Hardmodes I'd consider Haste to be slightly better until you reach around a 1000 and the "DR" really starts to kick in for real. For example my Penance is currently at 1.5 sec and casting up a PW:S for Borrowed Time which is 25% haste only decreases it with 0.2 sec or so. Getting "GCD capped" would be considered valueable, but is it really worth sacrificing so much to get there, you're talking hundreds of spellpower, int, spirit you'd have to sacrifice to gain those 0.1 seconds which you, more than 50% of the time don't really need?
02/10/10, 12:40 PM   #30
Squeakster
Piston Honda

Dwarf Priest

Eitrigg
 Originally Posted by kamaia The spellpower calculation was very back of the envelope and I'd like to do it "correctly" at some point. I thought I made that clear in the post, but re-reading it now I see I should have been more explicit about this. I made two assumptions in coming up with the 33 SP -> 0.48% ratio. The first one, which is probably not very big, is that I ignored the boundary cases where a non-overheal gets turned into an overheal. The second is I ignored the base heals of spells, so I simply said that 1% increase in spellpower would increase all non-overheals by 1%. 52% of the direct heals in the parse had some overhealing, and I looked up the disc priest from the parse on the armory and he had ~3300 sp unbuffed.
That makes sense, thanks for the explanation.

-----------

 Originally Posted by mutagen Before you go too far down this path you might check out shaman_hep and the accompanying EJ thread, It appears to be fairly well regarded and may even serve as a decent starting point for a priest tool.
It appears that the Shamans have beaten us to the punch. This is exactly the kind of thing I was thinking about, thank you for calling our attention to it Mutagen.

The page for shaman_hep states, "This program is written in Perl and needs a Perl interpreter," so I guess we need someone knowledgeable in Perl to adapt it to Priest healing. I wonder how difficult it would be? I would imagine it wouldn't be too bad, just changing spell names and some of the variables. Although I bet parsing absorb effects would require some work.

 Elitist Jerks On the Value of Haste