Talk:Hero behavior/Archive 02

Study failure (attributes)

Wooow people have really missed AI observation. It is stated that heroes and henchman share AI, which is fine but there is a third element that needs attention. Besides these two allies, players also have to consider foes, may not be the appropiate page but I just found this one and went to check for example Troll Unguent (page) as I have witness enemies using it while heroes with Wilderness Survival 12 dont use. The difference I noticed was that the foe had higher level than 20, cant remember which one I saw and its level but im awared of this. I have never tested a hero ranger with WS runes as only use Beast Mastery or Expertise. An AI can use a skill or not ALSO depending on the attribute points. This factor cannot be missed out. As in friendly AI can also elevate beyond 15 points. Yoshida Keiji talk 20:55, 2 January 2012 (UTC)

By the way, while checking "preview" Im getting the following message but I dont really know how to:

"Warning: This page is 54 kilobytes long; some browsers may have problems editing pages approaching or longer than 32kb. Please consider breaking the page into smaller sections."

Yoshida Keiji talk 20:55, 2 January 2012 (UTC)

If you are saying that there are three forms of AI behavior, I agree (probably, it's more like five: add allies and summons). In theory, they could all have different code for pretty much anything. In practice, I suspect that are mostly the same (except in scripted situations). I'm not sure why your heroes aren't using Troll Unguent, but (in part) I don't put it on hero skill bars b/c they do use it...at the wrong times (e.g. while they are standing next to a foe that keeps interrupting them). Any kind of testing has to consider other priorities and threats. Specifically, I'd put the White Mantle Scout build on e.g. Magrid and see if she uses it the same way the Mantle does.

If you want to share the build your ranger hero is using, we can try to replicate your results and see if it's a never use at WS≈12 or if there's something else going on. (If it's never, that sounds like a problem with the AI generally.) – Tennessee Ernie Ford (TEF) 00:44, 3 January 2012 (UTC)

I am still a noob wiki user and there are many things that I need to learn with coding and formating stuff but I made a scrap in my sandbox as how Hero AI behaviour should be studied which is the way I always did since NF release. User:Yoshida_Keiji/Sandbox Maybe this should also be addressed in a formating talk page but it starts here. Now to explain the reasons: I will put the example of GoLE I was totally upset and even rofl-ed when I read what others posted. The observation level was very poor as to just come and say...Hey I used on my hero monk bar and didnt work as I wanted so its bad skill...and then crossed out... Which in my opinion lacks seriousness...If the current page was all "analised" that way...I feel like we should probably start all over from cero.

About replicating foe builds thats something Im used to as well, I started studing like that after getting my first few elite skills as attempted to copy bosses when I was still newby. For the Troll Unguent talk I think my Sandbox has the answer for you.

And by the way... I think you need to make another archive: "Warning: This page is 40 kilobytes long; some browsers may have problems editing pages approaching or longer than 32kb. Please consider breaking the page into smaller sections. " Yoshida Keiji talk 07:08, 3 January 2012 (UTC)

Researching and documenting AI behavior is more difficult than you make it sound. Your template includes three columns labeled, observations, but they are conclusions based on observations. (I'm not doubting that you did the requisite testing, but, as you note above, many people jump to conclusions from small samples and a narrow range of test subjects.) A good reporting template would include the build used, the number of iterations, the variations in target foes, other heroes present, and so on. I think that's well beyond the scope of this wiki. I think even the simple presentation you have offered might not be used as well as you think (based on how I see the drop-rate data tables populated — that's much less complicated behavior and there's still trouble getting accurate data).

What you can do (as others have done) is to record your observations in your user space and report conclusions here (or in the relevant article). I don't think anyone has a problem with you (or other researchers) removing claims that contradict your research or that you cannot replicate...as long as there is a link to the methodology/data.

What I'd like to see happen with this page is that we remove the details of specific skill usage and focus on setting people's expectations of how the AI works. Instead of saying GOLE works like this and Leech Signet like that, we should offer the general guidelines: heroes will not use some skills unless they are spec'd high enough; they are less likely to cast powerful/useful AoE skills if there are only 1-2 targets; or they over-prioritize energy efficiency at the expense of casting life-saving skills (e.g. GOLE etc). This will turn the page into something that can be read/understood in a reasonable period of time, while we can include the specific examples elsewhere (a subpage for reporting to ANet and the relevant skill pages, where the specifics are more useful).

(Re: arching. The wiki warning is over-conservative: any 2012 browser that fails to be able to handle edits of ~50kb has other issues. I believe all of the topics remaining on the page are either current themselves or relevant to current discussions; they should not be archived just yet.) – Tennessee Ernie Ford (TEF) 08:46, 3 January 2012 (UTC)

Okay...so...if Im following you right, this would mean for me to create another page inside my user space. Is there a limit users are refrained from? And if it is common among users, which is the common name given to such space? The only I knew so far was Sandbox to test scraps. If anyone can help me also...my navigation bar will face trouble and I dont know how to handle such codings yet as it currently lists my characters using lot of horizontal space. Now returning to topic, if there is anyone who is used to "record observations" in user space, let me know so I can see how others do. Or if anyone reading this knows "users who research", point them for me please.

In user page for research I shall be allowed to post my builds and hero builds. Iterations. Explorable Area/Mission of testing. Team. Hero set-up (runes/weapons). For then later on arrive to specific skill pages to report conclusive data. Leaving this Hero AI behaviour page only for general overview. Yoshida Keiji talk 09:56, 3 January 2012 (UTC)

There isn't any limit to the number of userpages you can create for yourself, believe me, TEF and some of the others have a lot! Basically, you can create any varied hierarchy of pages you want, almost like folders inside Windows. The easiest way to create new pages would be to make a link for yourself within your namespace and then edit the empty page. For example, I created my character's page by typing out the link User:Farlo/Wreaking Havok, following the link, and then bam! you have yourself a page. If you'd like a generic sandbox to mess around in, just click this link → User:Yoshida Keiji/Sanbox ←, and start editing. As long as the pages start with "User:Yoshida Keiji/", they're in your user space and you can name and do whatever you want with them; there's no formhat to follow as long as it's prefixed by your username. All the normal wiki-code works on these pages as well, so you can setup templates, copy normal pages to mess around with, or whatever your heart desires. I hope this was enough info to at least get you started, and good luck!

If you want more help with your userpage, leave a message on my talk page so we can keep the discussion off of here ;) ~FarloTalk 11:08, 3 January 2012 (UTC)

User:Manifold has a number of research articles (linked from his home page; I think you'll be interested in some of his data). Also User:Judas has done some interesting work on Armor Ratings and the benefits of using shields/foci without meeting attrib-reqs. User:Tub also does excellent work (although I think, like me, he tends to post observations on the relevant article's talk page. In addition, there are any number of projects for collective research efforts. I do have a couple of research-like pages, but they are more along the lines of collating existing wiki data and putting it into a one-stop-shopping article.

Among the advantages of starting in your user space: you can fully control the form-vs-content and the direction of the research. Starting a wiki project is good if there are more than 2-3 people engaged in the same effort, but you'll notice that anything that takes hours and hours of effort tends to die down all-too-quickly (even those that involve bare facts and little disagreement). When I want other people to join in on my personal efforts, I add a note so that they know it's okay to edit in my user space. – Tennessee Ernie Ford (TEF) 20:56, 3 January 2012 (UTC)

As you guys may already know, AI skill usage Project was born from this. Im starting to make the first steps. I have seen that the "split suggestion" discussion hasnt moved foward at all. In the meantime, I would like to ask if (rewardless of the current inconvenience) we could sort alphabetically the skill listings (which I know, they shouldnt be in the same page of general behavior, but for now...). This is because from the Project... I find the need to double check skill pages (under study subjects) and also this section to make edits simultaneously in case we (or I ...so far...) change the note section due to research findings that prove different to current statements. Would be of great help here and in the future split section to have all skills sorted. Yoshida Keiji talk 16:15, 26 January 2012 (UTC)

A rare official comment about AI tendencies

“

The hero/henchman AI should be able to use their doublecast spells but they do tend to like to use them super-optimally, meaning they usually won’t use them unless they’re reaping most of the benefits of the skill. With Double Dragon, for instance, the AI might not use the skill unless both fire rings will hit enemies.

— Zack at the official forums

"[They]...like to use them super-optimally" probably explains a good deal of hero behavior for other types of skills, too. In Zack's example above, a player might use DD even if one of the fire rings would have no impact, since both caster and target-ally could still inflict burning. – Tennessee Ernie Ford (TEF) 19:38, 6 January 2012 (UTC)

Do heros ever use doublecast skills on allies, i.e. minions/summons/etc.? Random Weird Guy 20:16, 12 January 2012 (UTC)

I think RWG asked this before the first DCS update, but they have done so for at least a month. It's fun to cast EVAS and watch it flame on when a hero casts Double Dragon on it. (I dunno if DD is the most effective skill in the game, but I have a blast watching heroes use it.) – Tennessee Ernie Ford (TEF) 05:00, 19 February 2012 (UTC)

This kind of thing will never make everyone happy. If you change the AI so that it uses doublecasts when only one target benefits, you'll get people complaining that they don't use it optimally. -- Hong 05:36, 19 February 2012 (UTC)

Campaign Differences

It's probably my imagination, but heroes seem to be more passive in Prophecies than in other campaigns. For instance heroes will often just stand there even with foes in agro range casting enchantments, healing themselves, or raising minions. Can't quite tell for sure, but it sure feels like there's a difference. Perhaps the henchies do this as well, but I don't use them much anymore. Anyone else experience this? Lord Flynt 05:18, 24 January 2012 (UTC)

I don't recall ever experiencing this. Where was this, who were you fighting, and how was your heroes' energy? Kinda need more to go on to see what the problem was - may be situational and not campaign based. Konig/talk 11:29, 14 February 2012 (UTC)

2/16 Update

← moved to Talk:Hero_behavior/Unexpected_behavior_Archive_1#2.2F16_Update

Split (Mark II)

I disagree with Falconeye's added split suggestions. Reason being is that we should first figure out how big certain aspects are. "Positive" or "good" behavior is, from my personal knowledge, a general thing and as such can exist on the main article - for instance, these would include "very good at interrupting because [...]" or "good at being a minion bomber" and so forth. Specific skill use can likewise cover both positive and negative in what could be a relatively short (for showing both) article.
On a second note, would it really be helpful to split the good and bad between different pages? How would the article benefit compared to specifying generals (both good and bad) and split specifics (again, both good and bad)? And for that matter, do we even really need to split? Some collapsing tags should settle any issue of length (other than loading time, but since we don't have tables and tons of images, we shouldn't have too much issue in that).
So again: disagree on the basis that splitting between positive and negative behavior is uncertain in length and I don't see it as being more helpful than the previous suggested split. Konig/talk 02:03, 17 February 2012 (UTC)

Assassin's Promise

← moved to Talk:Hero_behavior/Unexpected_behavior_Archive_1#Assassin.27s_Promise

"I Will Avenge You!"

← moved to Talk:Hero_behavior/Unexpected_behavior_Archive_1#.22I_Will_Avenge_You.21.22

Blackout

← moved to Talk:Hero_behavior/Unexpected_behavior_Archive_1#Blackout

Ebon Dust Aura

← moved to Talk:Hero_behavior/Unexpected_behavior_Archive_1#Ebon_Dust_Aura

Ritual Lord

← moved to Talk:Hero_behavior/Unexpected_behavior_Archive_1#Ritual_Lord

Disrupting Dagger

← moved to Talk:Hero_behavior/Unexpected_behavior_Archive_1#Disrupting_Dagger

Destructive Was Glaive

← moved to Talk:Hero_behavior/Unexpected_behavior_Archive_1#Destructive_Was_Glaive

Healers?

I've long wondered if the Healer AI is purposely nerfed; perhaps to promote Players to take the role on themselves; but I am sure I am not the only person to have noticed that if at the end of a battle my PC is taking a dirt nap; my Minion Master will always exploit my corpse before my healer 'thinks' to Rez me; or that in a battle the time a Hero/Hench Healer takes for any spell is padded by the little Dance they do prior to activating any skill; which say a Forgotten Monk does not share. No the Mob AI is clearly more efficient, and as the complaints about Alesia and company go back to the very beginning of the game; one can not help but suspect the class has been purposefully nerfed. With the Launch of GW2 just days away I know minimal effort will go into maintaining GW1; but removing the little Dance from Alesia's command sequence would likely not involve a huge patch and would radically increase her efficiency. So Please do. ... --69.255.123.188 11:31, 1 July 2012 (UTC)

Need more details and less abbreviations:

• Which is the Minion skill? What weapon does your Minion Master (upgrade mods) have? Does your necro hero have Bloodstained Insignia? What resurrection skill does your healer have? What weapon does your healer have (upgrade mods)?

• What is HH? Healing Hands? Holy Haste?( Hero/Henchie Sorry I thought that was a pretty Standard abbreviation.)

See also: Talk:Hero_behavior/Unexpected_behavior#A_note_on_Hero_AI_vs_NPC_AI

Yoshida Keiji talk 12:36, 1 July 2012 (UTC)

Re: Aura of the Litch (2Sec)/Vampiric Horror or Shambling Horror (3 sec) Vs Resurrect (5 Sec) but the MM always exploits the corpse before the Monk even begins her Wind-up dance. Assuming the MM should not always be idle or at range at the end of a fight, she should not always be first; but she is it's every consistant; in fact you can very often count one or two sec off after the corpse is exploited,

I don't think she's waiting on the Necro, because I have noticed the same delay if a PC Toon or Hero needs healing or a condition removed; there is always a long pause; if I have that hero bar up it's definately quicker to micro the healer than wait on the AI to react to the situation. Neither the Healer or Necro has a significantly better staff nor overall energy pool. Obviously one could only be certain about these things if one could view the AI code; but I strongly suspect a delay has been built into the Hero Healer AI myself. --The preceding unsigned comment was added by 69.255.123.188 (talk) at 18:18, 2012 July 1 (UTC).

Went to Isle of Namless, brought a hero with only Animate Bone Minions, and a hero with only Resurrect Chant. Blood is Power-ed myself to death (casting on Master of Paths). They both used their skills within a quarter second of me dying, simultaneously as far as I can tell.

Heroes prioritize healing over resurrecting, which is good. Your MMs prioritize making minions, which is good. I don't see any issue. Manifold 19:18, 1 July 2012 (UTC)

Or maybe it's bad, but either way, it's a clear and consistent design choice rather than an unambiguous bug in the AI. That makes your ideas above suitable for a formal suggestion, rather than a bug note. – Tennessee Ernie Ford (TEF) 19:35, 1 July 2012 (UTC)

Re I will do some testing with Resurrect chant, I was not suggesting it so much as 'Bug Note' as an observation of AI behavior; I do concurr I think it probably reflects and early design choice (7 years on who knows if the Current Devs would be aware of it; assuming I am right in my observation.)

I do know alot of players dont even use Monk Healers relying on Rits or Necro's instead; because as I said above the monks have been notorious from the beginning. If I may suggest those interested do a little play testing work slowly through an area and watch the Monk's response time in that moment with the dust is settling; I think you will find I have a point. --The preceding unsigned comment was added by 69.255.123.188 (talk) at 20:39, 2012 July 1 (UTC).

In the meantime, if you want a faster resurrection skill for monks, try Unyielding Aura, you will see monks do a lot better with that one. Yoshida Keiji talk 20:56, 1 July 2012 (UTC)

Thanks Much I will. --The preceding unsigned comment was added by 164.86.207.250 (talk) at 22:57, 2012 July 1 (UTC).

I've noticed the same thing, within milliseconds of my death, I'm rezzed as a minion before the monk even thinks of using a rez skill. And I do use Unyielding Aura, and I've seen that monks seem to take a long time to re-cast UA when there are multiple deaths. I've even had to micro UA because the monk refuses to cast it when there are 2 or more dead party members. *Malganis Frostmourn* 99.48.31.204 00:19, 2 July 2012 (UTC)

Okay...now here we have 3 different IPs but I assume the first two are the same person. If we are going to keep this talk, I suggest you guys register an account here so we know who is who.

Other already documented observations are: Hero_behavior#Resurrection and Unyielding_Aura#Notes. Depending on the case, expecting a monk to rez in mid-combat might no suit the moment as during the resurrection cast time...another party member could need healing...or your party will end up with more casualties. If you have two heroes and your MM animates a minion and then another hero, a monk, resurrects you it ain't a bad thing, because you returned to life and your team got 1 more friendly. And yes UA behavior by AI is true. Yoshida Keiji talk 10:40, 2 July 2012 (UTC)

Healing and Protection

"They prioritize the ally with the lowest health, regardless of importance to the party. This makes them unreliable to protect mission-critical non-party NPCs (e.g. Prince Rurik or Master Togo)."

"They prioritize the ally with the lowest health and will not protect goal-critical non-party NPCs (e.g. Prince Rurik or Master Togo) unless the player targets that character first."

• Not many people know to heal NPCs. The ones who do know, are well aware that heroes can target non-party friendlies as well.

---

"When an ally is near death, NPCs will sometimes favor skills with an effect-on-death over healing, e.g. Death Nova instead of Heal Other."

"NPCs will detect near-death allies and make proper use of appropriate skills e.g. Divine Intervention."

• That was a noob input by some Discord-way (a.k.a. brainless-way) play. Change your noob necro hero with Death Nova for a monk with Divine Intervention and you will then see how a hero will give priority to this skill over other healing skills.

---

Yoshida Keiji talk 10:32, 10 July 2012 (UTC)

I disagree with both of these changes. It's not up to us to decide the discordway is bad (in fact, it's extremely effective in a huge number of situations, including the AFK Glint's Challenge farm). Similarly, you cannot force a hero to target an ally without constant micromanagement: you may lock a hero onto a foe, but you cannot lock a hero onto an ally. The AI is entirely unreliable on its own for ensuring the survival of mission-critical allies and I think it's misleading to suggest otherwise. – Tennessee Ernie Ford (TEF) 15:02, 10 July 2012 (UTC)

The current description needs to be changed, I would do it right away but I will discuss it further first to avoid reverts/undos + ECs (also im hungry). What we need to state is that AI does properly use near-death skills. What players should be warned about is what "near-death" skill are incorporating to a build. In the Discord-way, it is entirely the player's fault. Because he/she assigned the wrong "near-death" skill to a character with healing skills. I find it extremely bad idea as a team build maker because this would mean, the user is expecting a Minion Master to heal...which is not job specific (although I am aware of the benefit due to Soul Reaping energy convenience). And that is exactly why I say Discord-way = Brainless-way. (I shall start a my-rants section + userbox for this.)

Regarding NPCs survival, without heroes: Yes you are right. But with heroes: If you need constant micromanagement, the answer is bonds (with the exception if game developer failures e.g.:Nolani Academy in HM). Yoshida Keiji talk 07:46, 11 July 2012 (UTC)

Ah, I see what you mean about brainlessway (I still disagree, but... your point is valid... and will make me think about it some more). So I think it's that players need to know two things:

• The AI will focus on near-death targets using near-death skills, but...
• They will not choose as expected when their build includes both heals and effect-on-death. (Death Nova: great on a minion, not so good when your UA is at 5 HP.)

I don't think there's anything wrong with expecting MMs to heal, but once players are aware of the AI's decisions, they (the players) should know better. I think this is comparable to including touch skills on a healer. Yes, heroes know how to use them and yes, they know how to stay out of aggro, but they will not choose dependably. This makes no difference when the team is winning, but makes all the difference in determining whether near-wipe becomes a wipe or not (same as with MMs who have Death Nova + heal).

Regarding the non-party member survival. My point is that the AI doesn't do a good job of prioritizing mission-critical allies without player intervention: if MOX has 20 HP and Rurik has 30, the AI will try to save MOX and not care that the team will wipe. Yes, you can use bonds, but that still requires micro (albeit, you can set it up pre-battle). The note, imo, should demonstrate that you cannot reliably depend on the AI to save the allies, regardless of whether any workarounds are better or worse, easy or difficult. – Tennessee Ernie Ford (TEF) 08:24, 11 July 2012 (UTC)

The problem about expecting a MM to heal is that it will be casting heals with what? Woe Spreader? Some of my default Olias between all my 10 character have it because its a 2k-3k green, but soon replaced with Death Magic staff with HCT 20%x2. A PvE mainstream player may not mind about it but any PvX player knows the disadvantages. That said, double function on a same character when 2 could sum simultaneos action is bad for me. I prefer my team to be rising minions at the same time its healing and not only one of two. Yoshida Keiji talk 09:15, 11 July 2012 (UTC)

Sure, that's why I stopped using dual-purpose MMs. But it's not up to us to impose our views on the wiki, but instead explain how the AI works well enough so that other players may draw their own conclusions (even if they end up with bad ideas/builds, at least we can say, "told you so").

Regarding kit: honestly, I stopped optimizing hero kit a long time ago (too big a pain to switch equipment every time I wanted to change builds). So far, I haven't found that it matters much, except in very narrow circumstances (for player builds, esp. PvP, it matters a lot). I take time to give spears to caster heroes for the Deep (to avoid infinite Chaos Storms), but Oggie heals just fine with a generic divine staff and Ollie pumps out 10-11 minions fast enough with a generic death staff. – Tennessee Ernie Ford (TEF) 09:30, 11 July 2012 (UTC)