League of Legends XXV: Who is your Summoner, and what does he do?

[Qwertystop]

What would happen is that that person would be hired on the spot, and I'd be serious impressed at someone who managed, by themselves, to build a complex, champion-specific AI for a game with 90+ champions while still holding down a different job and/or not going mad.

Especially since I don't think Riot has made their code public.

Actually, that might mean that the person would be arrested on the spot.

[Djinn_in_Tonic]

Actually, that might mean that the person would be arrested on the spot.

Not sure on that one...the fact that LoL Replay both exists and works implies that someone has built something to interface with (and to some extend copy or allow to be copied) Riot's code, and LoL Replay still works. So I'm not sure Riot would sue, although they'd probably be within their rights to do so.

[Darth Mario]

Not sure on that one...the fact that LoL Replay both exists and works implies that someone has built something to interface with (and to some extend copy or allow to be copied) Riot's code, and LoL Replay still works. So I'm not sure Riot would sue, although they'd probably be within their rights to do so.

Riot wouldn't sue.

They'd hire the person on the spot.

[Arbitrarity]

Yeah, though the only part of designing an AI system without working with Riot that would be really sketchy would be interfacing properly with the game and client. I'm not even sure you could do that without somehow reverse-engineering large parts of their codebase, and it would be pretty impossible to test. But yeah, you don't sue people who did multi-month software projects independently to improve your product.

Also, LolReplay isn't very technically complicated. It just records packets and feeds them back into a slightly modified client. I suspect they use a method similar to UI modding, though I have no idea how people actually do that for LoL.

[Delusion]

Tried Jungling with Riven again on custom game. (last attempt was at level 18 or 20 I think) And it went well 30/0/2, but it was a bot game so it doesn't relly count.

I think it was either the new Masteries or the fact that I went cloth + pots instead of Vamp scepter that allowed me to not get killed by the mobs at level 2. Or that I have a lot more runes. Or all of the above.

Went Mini golems -> Wolves -> Wraiths. Is that totally horrible path?

Oh and how do I build Zilean? Having lots of fun with him but no idea if my build is at all effective.

[ArcanistSupreme]

Honestly? It takes a long time to build an AI, especially one with as many variables as you guys want to see. Especially so if the AI isn't meant to be perfect...a perfect AI is much easier to build than one that it just *good.* If they focused on the AI improvement you want, we'd see a marked decrease in all other content, including champions and patches (or an incredibly long time before any AI improvement could ship out).

Note: The following statements are made by a person with super limited programming experience (as in this person has made programs for TI-83+)

Wouldn't putting an RNG into the calculations go a long way toward toning them down? It would definitely take some tweaking, but that would go a long way toward making them seem more human. An RNG could also be attached to their reaction speed.

I don't know about you, but I don't value AI that highly at my level of play. It works for beginners, and I can always solo queue or ranked queue if I want a challenge.

Well, there's "challenge" and then there's "trying to win a game 4v5 with two feeders." But seriously, I play bot games to learn new champions or if I only have half an hour to play LoL and want the 150 extra IP. Why not make that half hour more stimulating?

What would happen is that that person would be hired on the spot, and I'd be serious impressed at someone who managed, by themselves, to build a complex, champion-specific AI for a game with 90+ champions while still holding down a different job and/or not going mad.

Then again, that person doesn't have to do it all by himself/herself. I'm sure that there are plenty of people out there that would love to get in on that project. Heck, I bet some people would do it for free if Riot gave them some of the code to play with.

I'd honestly rather not, and feel quite the opposite. The new (and old) tutorial serves its function pretty adequately in demonstrating the basic skills; "Don't tower dive." "Last hitting is important." "The brush hides you from sight.". A couple of those take lots of practice to get good at, especially across different champions, but I really am starting to dislike the dependence on bot games...I mean, it's kind of the point of the game anyway, and I made it to where I am just fine without bots. Most of the suggested changes to make them "harder" could just be avoided by making people play games against actual people in the first place.

I'm not advocating for bot games to replace real games. I'm advocating for bot games to teach better habits that, quite frankly, aren't even taught in real games before a certain ELO. The tutorials don't teach players the importance of last-hitting, as shown by the abysmal creep scores in 98% of normal games/ranked games at my ELO. And when I play bot games to get a feel for a new champion without making 9 other people suffer, I'd like to get an actual feel for how much damage they can dish out and a slightly better idea of how to play that champ.

tl;dr
Play real games for a real challenge. But that's no reason not to make bot games better.

Yeah, though the only part of designing an AI system without working with Riot that would be really sketchy would be interfacing properly with the game and client. I'm not even sure you could do that without somehow reverse-engineering large parts of their codebase, and it would be pretty impossible to test.

For a good part of my life, I thought that games just magically appeared on shelves in game stores and never really thought about the process they had to go through to get there. Just out of curiosity, how big of a step is it to go from mathematical formulas and instruction on how to implement them to game-specific code?

But yeah, you don't sue people who did multi-month software projects independently to improve your product.

That would be a little tacky.

[9mm]

Riot put up a massive resource map when they changed the login clients so modders could still mod stuff. However they tightly control what is on your machine that can be modded.

[Djinn_in_Tonic]

Riot wouldn't sue.

They'd hire the person on the spot.

That is what I initially said, yes. See the previous page of this thread.

Wouldn't putting an RNG into the calculations go a long way toward toning them down? It would definitely take some tweaking, but that would go a long way toward making them seem more human. An RNG could also be attached to their reaction speed.

Yes, it would. But then you need to figure out how the RNG work for every case. Is it their reaction time? The direction they move? The distance they move? The speed or order in which they use their abilities? The amount they can under or overestimate their damage / the opponent's health / the opponent's damage / their resist / their opponent's resist / the damage minions deal / tower range / ...and so on?

There are simply to many variables to account for. Hence why human-level AI is all but impossible: you can't really match the semi-randomness of the human element in anything even close to resembling an accurate manner.

Well, there's "challenge" and then there's "trying to win a game 4v5 with two feeders." But seriously, I play bot games to learn new champions or if I only have half an hour to play LoL and want the 150 extra IP. Why not make that half hour more stimulating?

Because, as I said, the amount of coding and bug-fixing required would equate to a huge decrease in Riot's productivity in other areas, and "better bots" isn't really that high on the player-bases priority list (I would imagine). Sure, it would be nice. But I don't think the time and expense would justify the results.

Then again, that person doesn't have to do it all by himself/herself. I'm sure that there are plenty of people out there that would love to get in on that project. Heck, I bet some people would do it for free if Riot gave them some of the code to play with.

Maybe. But companies in general are extremely loathe to do that, for a number of ethical and legal issues.

For a good part of my life, I thought that games just magically appeared on shelves in game stores and never really thought about the process they had to go through to get there. Just out of curiosity, how big of a step is it to go from mathematical formulas and instruction on how to implement them to game-specific code?

From very little to incredibly byzantine and complicated, depending on the level of interaction with other things that must be accounted for. Giving a bot the ability to pick a primary target based on, say, low health (80% priority) and distance (20% priority) is relatively easy. The more variables and interaction you add, the more complex the process. A bot who can analyze a teamfight and pick a target based on location, spells used, current health, total resistances, current and/or likely target, amount of CC, amount of burst, amount of sustain, summoner spells on cooldown, estimated masteries and/or runes, allies to save and/or protect, map objectives to claim, bushes and line of sight to exploit, ward placement, knowledge of teammate and enemy positioning...and then places a changing value on all of those statistics to determine the best course of action every second while not processing so quickly that it can't make a decision because data changes every time before it can cast any spells? You can see how this gets complicated.

[Daverin]

Instead of actually contributing to the discussion, I just want to say how awesome it is you actually used the term byzantine in a conversation.

[Arbitrarity]

For a good part of my life, I thought that games just magically appeared on shelves in game stores and never really thought about the process they had to go through to get there. Just out of curiosity, how big of a step is it to go from mathematical formulas and instruction on how to implement them to game-specific code?

In the framework I just described, you'd basically need a whole bunch of hooks into the current gamestate to implement pretty much anything. Optimally you'd be able to run at a server level, which should give you full access to game data (i.e. current health of creatures, exact distances, cooldowns, etc). In that case, a lot of these checks could be done really directly, by referencing the relevant object properties. Without direct access to their codebase though, this is impossible. We'd need their server code, which isn't distributed anywhere.
I suppose at server level, functions like "Cast this spell targeted at this location" could be done easily, though it would be better practice to spoof a client instance to do that with, as that means it would go through the usual player pipeline, which prevents you from getting bugs like "Oh, Annie Bot can cast Tibbers globally"
So you'd have your coordinating Team object, and 5 instances of championAI objects with champion-specific properties. Each championAI would have a spoofed client in order to interface with the game correctly, though it should only really be used to send requests. They might have another implementation for getting bots to send commands, which has less overhead; I'd defer to using that, since all I need to do is send. You'd need to instantiate those on game start, linked to the specific game, though that's mostly basic implementation detail that you totally can't do without understanding a decent chunk of Riot's architecture, praying they have a good development platform (hint: this is improbable) or recycling whatever they use right now for AI. You'd also have to figure out what sort of time for a cycle you want the AI to run at (this acts like minimum possible reaction time, but is really a performance concern).
Mostly though, this would act like a very seperate entity from the game. The server would just have a Team object sitting around polling it for data, with a bunch of championAI's asking the Team for data/instructions, then acting by sending requests like any other player to the server (albeit with 0 latency). The hooks into stuff like "cast this spell here" could then be pulled from the client's standard messaging, which wouldn't be that bad; there aren't that many different actions in LoL (qwerdfb, shopping, camera movement, and we don't care about the camera). The painful part is getting the interfacing abstracted away so that instead of saying "Shen casts E targeted at enemy under these conditions", you can do something like "tanks CC priority target in teamfights", then fill in what CC the tank has, and its properties. You could also try to develop some sort of abstract way to read champion properties, but I think that's more trouble than it's worth; while it would mean big reworks wouldn't require AI changes, and new champions wouldn't require much manual work, the amount of effort needed to prevent an hour's work every few weeks doesn't seem worth it. You need to abstract it a BIT, in order to maintain DRY, (i.e. don't have to tweak champion AI every time a hero gets a minor change like a range nerf), but beyond that, a reasonable amount of champion-specific properties would probably be OK.
I'm also no longer sure if I want specific champions to be instances of championAI, or inherit from it, then instantiate themselves. I think inheritance is better, as it would consolidate champion-specific functions into a single file.

Actually Djinn, the problems like "what summoner spells are up, positioning, burst/sustained damage" aren't that bad. Technically, making them run at a reasonable pace might be difficult, but from a coding perspective, that's just scripting on top of a pretty simple framework (or modifying the framework, as the case may be). Mind you, I suspect it would be a lot easier for bots to cheat on some things (does this guy have Smite up? Does he have Flash up? What masteries does he have? Runes?) than not, and getting some corner cases would probably be a pain, but probably doable. I personally think the hard part would be initially making the framework you could work those sorts of incremental improvements into; without seeing how Riot coded that (judging from bot post-game behavior, the answer is they cheated) I can't really tell how much of that work is done. The lack of incremental bot improvement and lack of bot variety says "not well"

[ArcanistSupreme]

There are simply to many variables to account for. Hence why human-level AI is all but impossible: you can't really match the semi-randomness of the human element in anything even close to resembling an accurate manner.

Fair enough. But better is better, even if it's not perfect.

Because, as I said, the amount of coding and bug-fixing required would equate to a huge decrease in Riot's productivity in other areas, and "better bots" isn't really that high on the player-bases priority list (I would imagine). Sure, it would be nice. But I don't think the time and expense would justify the results.

Probably not. But nice things are nice.

Maybe. But companies in general are extremely loathe to do that, for a number of ethical and legal issues.

Right, but I'm sure some kind of contract could be drawn up. Then again, time/resources and all that...

From very little to incredibly byzantine and complicated, depending on the level of interaction with other things that must be accounted for. Giving a bot the ability to pick a primary target based on, say, low health (80% priority) and distance (20% priority) is relatively easy. The more variables and interaction you add, the more complex the process. A bot who can analyze a teamfight and pick a target based on location, spells used, current health, total resistances, current and/or likely target, amount of CC, amount of burst, amount of sustain, summoner spells on cooldown, estimated masteries and/or runes, allies to save and/or protect, map objectives to claim, bushes and line of sight to exploit, ward placement, knowledge of teammate and enemy positioning...and then places a changing value on all of those statistics to determine the best course of action every second while not processing so quickly that it can't make a decision because data changes every time before it can cast any spells? You can see how this gets complicated.

Yeah, but I'd settle for starting super simple and then every know and then stepping it up a little with a new patch. Based on the ferocity of the Riot Wars, I think it's safe to assume that some employees have some down time in which they could work on other projects.

...I just want to say how awesome it is you actually used the term byzantine in a conversation.

I concur.

[TechnOkami]

Instead of actually contributing to the discussion, I just want to say how awesome it is you actually used the term byzantine in a conversation.

I'll second this.

[Djinn_in_Tonic]

Instead of actually contributing to the discussion, I just want to say how awesome it is you actually used the term byzantine in a conversation.

It's such a great word that I feel obligated to use it.

Plus, I do have a quota of awesome that I have to live up to on a daily basis. Otherwise my funding gets cut.

[Daverin]

It is indeed a great word! Plus, for me, I have the added bonus of considering the Eastern Roman Empire as one of my favorite subjects of history. So, yeah, the term byzantine is twice the fun for me!

Meanwhile, obligatory complaint that new skin does not appear to be up yet.

[Lix Lorn]

Plus, I do have a quota of awesome that I have to live up to on a daily basis. Otherwise my funding gets cut.

From what I've seen of you, it seems more likely that you have a maximum awesome count per day, and get told to tone it down to not outshine everyone else.

[Darth Mario]

Plus, I do have a quota of awesome that I have to live up to on a daily basis. Otherwise my funding gets cut.

Which is to say, people stop donating to the mumble server.

In other news, donate to the mumble server! It's a cool thing to do and contributes to ongoing awesome.

[ArcanistSupreme]

Programming stuff.

So, if I'm understanding this correctly, what you are proposing to do is create an overmind for the bots that controls and coordinates them? That just sounds cool.

Of course, the easiest way to improve bot performance is to add a line of code that says, "If laning phase is over, then group together and under no circumstances should you chase an enemy into the jungle by yourself."

On a completely unrelated note, did anyone else notice that Chaox built a Tiamat in the final game against Epik yesterday? Obviously, this means a totally viable item because it saw tournament play.

[TechnOkami]

I'm curious: Has anyone bought Blood Lord Vladimir yet?

[Daverin]

Yep, the moment I saw it.

Actually is pretty cool. The enhanced particles really add to Vlad, and his new VOs are pretty sweet (and lack the puns, for goor or ill.) His dance is basically a reference to Fuji Balls.

Also, when Q'ing, his cape has this awesome glow that fades after about a second. And he now trails blood as he walks (and not just the blood... ball... thing he carries around, like all around him.) Other little things like that which serve to enhance the skin's appeal.

As a whole, it is a complete legendary skin. Admittedly, it is not some hilarious reference or else a particularly famous concept like Soul Ed... I mean Demonblade Tryndamere, but it definitely shows the work put into it.

Speaking of Vlad, I am convinced that one of the best ways to go about increasing his damage output is a Lich Bane, since he pretty much will be proccing spells around 2.5 to 3 seconds, and if you space them out it means you can basically be like a ranged, mage Riven. The only thing is I cannot figure out when it needs to be there (i.e. when he has enough survivability and the AP route just is dipping off,) and what to replace it with. I am thinking maybe it could come after Deathcap, except then that delays Rylais, which is now very useful for him to have. My gut instinct as far as replacement goes is that it could replace Spirit Visage and I could use Blue Pot for CDR, although the health and increased regen is an extra bonus for Vlad. Still, with a WotA (I so would've chosen Gunblade pre-nerf considering Lich Bane), near 3.6k health, and equal MR to SV, I think Vlad would appreciate the damage more than the survivability at that point in the game. The only other thing I'd even think of replacing is the Voidstaff/Abyssal, but... no. That's just a bad idea.

[ex cathedra]

Speaking of Vlad, I am convinced that one of the best ways to go about increasing his damage output is a Lich Bane.

But... why? RQEQE etc hardly leaves you time to wait for an awkward, slow auto-attack windup that doesn't A: heal you or B: do AoE damage. You could spend >3000 gold on an additional 1.0 AP ratio ability... or you could use Q instead.

[SlyGuyMcFly]

Actually is pretty cool. The enhanced particles really add to Vlad, and his new VOs are pretty sweet (and lack the puns, for goor or ill.) His dance is basically a reference to Fuji Balls.

Ill, most certainly for ill. The new particle effects are bloody awesome though. I'll give 'em that.

you know I had to do it

[Daverin]

But... why? RQEQE etc hardly leaves you time to wait for an awkward, slow auto-attack windup that doesn't A: heal you or B: do AoE damage. You could spend >3000 gold on an additional 1.0 AP ratio ability... or you could use Q instead.

Err, yes, yes it does. That's the point. No, you're last spell used will NOT be up before the time it takes for one autoattack, even for Vlad's rather awkward attack, AND considering 40% CDR. Of course, you can just burst out the Q and E, but I think that getting in 2 Lich Baned autoattacks is more than worth slightly separating them.

And, getting an additional 1:1 AP ratio ability that fits inbetween Q and E instead of spending more money on AP on two abilities with currently fairly poor AP ratios? Err yes, yes I will pick that. At least so long as I am looking for damage.

[ex cathedra]

Err, yes, yes it does. That's the point. No, you're last spell used will NOT be up before the time it takes for one autoattack, even for Vlad's rather awkward attack, AND considering 40% CDR. Of course, you can just burst out the Q and E, but I think that getting in 2 Lich Baned autoattacks is more than worth slightly separating them.

The faster you cast them, the sooner they can cool down... If you refrain from using your Q/E at all you lose a lot of potential damage. Additionally, do keep in mind that E is .45 in an AoE and LB is 1.0, auto-attack reliant, and single target. Due to the proc cooldown, I'm nearly certain that waiting for a lichbane proc between each and every cast of Q and E is a terrible idea. Q->proc->EQ->proc->EQ->proc at least pumps out a decent amount of spells.

If you absolutely must sell your SV for something with more lategame presence I'd select Hourglass or even GA nearly every time.

[Kairaven]

my current vlad test build is health and AP with a touch of Attkspd. only defense comes in runes, masteries and a bit of MgRes from whatever that armor that gives your 15% increased heal+regen rate.

-------------
I been having problem with my chasing game when on Ranged AP. the problem it seems is that usually if I burst someone down with a combo, and they survive, they are usually at range, and melee and AD range all have movement/dash skill/item that makes me unable to chase them down (unless flash of course, but really, that's only available once per 3 min or so) the only AP movment speed item that can be considered are Lichbane and FoN, and both of those are far fringe item for a range AP.

any help on how I chase down for the kill?

[Daverin]

The faster you cast them, the sooner they can cool down... If you refrain from using your Q/E at all you lose a lot of potential damage. Additionally, do keep in mind that E is .45 in an AoE and LB is 1.0, auto-attack reliant, and single target. Due to the proc cooldown, I'm nearly certain that waiting for a lichbane proc between each and every cast of Q and E is a terrible idea. Q->proc->EQ->proc->EQ->proc at least pumps out a decent amount of spells.

If you absolutely must sell your SV for something with more lategame presence I'd select Hourglass or even GA nearly every time.

Nope, just checked again just to be 100% positive. It runs extremely smoothly, with: spell, attack, other spell, attack, first spell up right after, attack, other spell up right after, attack, rinse lather repeat. The only way it would actually cut into your casting is if the first spell was up before the second autoattack, which it is not. In other words, not a single beat is missed. You still get your spells out the moment they come off CD, while getting a new, lovely 1:1 AP ratio ability that does not delay your next round of casting twice in a cycle.

[SlyGuyMcFly]

any help on how I chase down for the kill?

Generally, Rylai's. Gives a respectable amount of AP, a nice chunk of HP for surviving stuff and, of course, dat slow. I build it on Annie every now and again when the other team is too tanky to burst properly or too slippery to finish off after Beardropping them.

I see it all the time on Cassiopeias and Brands.

[ex cathedra]

any help on how I chase down for the kill?

Which ranged AP champs are you talking about?

Anyways, you simply probably shouldn't try to burst them until you're certain that you'll get the kill. Until then, just harass with your safest/most efficient source of damage. Once they're low enough, CC them and then let your burst+ignite finish them off.

[Daverin]

Which ranged AP champs are you talking about?

Anyways, you simply probably shouldn't try to burst them until you're certain that you'll get the kill. Until then, just harass with your safest/most efficient source of damage. Once they're low enough, CC them and then let your burst+ignite finish them off.

This sounds about right for most mages (not entirely sure what you mean by Ranged AP, but from your post I assume you mean bursters.) If you are doing this, then it is pretty much you just need more practice with the champ/experience to judge when that critical point is.

[ex cathedra]

This sounds about right for most mages (not entirely sure what you mean by Ranged AP, but from your post I assume you mean bursters.) If you are doing this, then it is pretty much you just need more practice with the champ/experience to judge when that critical point is.

Kairaven's words, not mine. Was using them for simplicity.

Nope, just checked again just to be 100% positive. It runs extremely smoothly, with: spell, attack, other spell, attack, first spell up right after, attack, other spell up right after, attack, rinse lather repeat

Mmm, in that case, you aren't getting a lichbane proc between every cast. Q->proc->E->no proc, as early as the first cycle, correct?

[Daverin]

Kairaven's words, not mine. Was using them for simplicity.



Mmm, in that case, you aren't getting a lichbane proc between every cast. Q->proc->E->no proc, as early as the first cycle, correct?

Oh I know it was at Kairaven, I just quoted your's because I thought you were right about it.

Also, I believe I did see the proc on each cycle. I will check again to be sure, though.

EDIT: No wait, my memory fails me too easily. I didn't actually get the Lich Bane that check, just went to see if the cycle fit. Either way, going to check.