Target Lines on Red Team

mcompany
Autonomous Entity
Autonomous Entity
Posts: 872

Re: Target Lines on Red Team

Post#21 » 01 Jul 2017, 02:48

First of all, copying the teleporting resource idea and ritter's aggressive sniper both are things that aren't aided at all by the red lines, but by observing the enemy behavior.

Second, for anyone's fire system, I'd image that it would be pretty hard to accurately copy. The reason for this is because for someone like me, I have 6 nodes per range to determine what order I attack enemies when I attack without aiming at specific stuff. How does one ever solve that with only lines? How does one ever know I'm not somehow having very specific targets? And the more complicated the AI, the harder it is to copy. Add on top of that, if you do manage to copy everything against you to act the exactly same way as when you faced the real AI, how do you know that it is actually as strong as the original and can't just only beat you (overfitting)? You can say all you want that some random might be able to easily copy your AI with some patience, but it simply requires WAY more patience than "some" and requires so much research that you are better off just making a better AI in the first place than copying. Which is why I tell you: please come up with an easy method to 1. correctly distinguish the difference between a 20- node AI and a 150+ AI, and 2. correctly copy a 150+ node AI to at least keep it at a similar strength of the original by ONLY using observation. Because I highly doubt that anyone could even perfectly copy my Mind Game AIs when both of them are less than 20 nodes, much less that you suddenly would have had Wozza's whole system if the lines were available sooner
Castlevania wrote:Copying something from community is not as convenient and reliable as copying from the opponent who just defeat you

Third, this statement doesn't even make that much sense because you not only get a full picture of what node goes where from a community screenshot, but you also usually get a full explanation of why each node goes where it goes. Meanwhile, to copy an opponent, you have to guess at what they are doing, why they are doing it, whether or not any of it is based on tags, and whether or not it is made to be map specific, and then on top of that you have to somehow figure out a way to use the information to make it useful in your own bots

So, do I think the lines makes copying easier? Quite possibly. Do I think it is enough that tells the enemy what is going on? Not really. And so far the only thing you are complaining about is a "what if" situation that has been proven to be possible for months that simply have never happened because it is not worth it

mcompany
Autonomous Entity
Autonomous Entity
Posts: 872

Re: Target Lines on Red Team

Post#22 » 01 Jul 2017, 06:43

Amongst the arguing, I forgot that I meant to do this:
Castlevania wrote:I also hope that there is a button to control on/off of target lines. They are not real things like bullets.

+1, although I already +1'd this which is basically the same thing, only slightly different:
TheGnoo wrote:Maybe just a button to switch between red and blue could be cool.

User avatar
Kanishka
Skynet
Skynet
Posts: 1421
Contact:

Re: Target Lines on Red Team

Post#23 » 01 Jul 2017, 07:51

Okay, I am quoting mcompany from the Telegram Chats... "You can only replicate the part of the Enemy's AI that was activated against you." And there's no way to be sure of all conditions and priorities.
Fixes break an AI more than bugs do. :ugeek:

Gladiabots Off-Topic Chat


My Stats: Kanishka_RN3;Kanishka_MiPad

pier4r
Skynet
Skynet
Posts: 3390

Re: Target Lines on Red Team

Post#24 » 01 Jul 2017, 09:40

The wisdom of mcompany captured by Kanishka
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

User avatar
Kanishka
Skynet
Skynet
Posts: 1421
Contact:

Re: Target Lines on Red Team

Post#25 » 01 Jul 2017, 11:12

pier4r wrote:The wisdom of mcompany captured by Kanishka


This. (Though I added a line. ^^)
Fixes break an AI more than bugs do. :ugeek:

Gladiabots Off-Topic Chat


My Stats: Kanishka_RN3;Kanishka_MiPad

Captain Spock (Castlevania)
Automaton
Automaton
Posts: 177

Re: Target Lines on Red Team

Post#26 » 01 Jul 2017, 17:18

Hey guys, if you try to argue about accurate copy, you misunderstand my point. The most common element that spread out through the matches are "features". I remembered once that a lot of players (include me) used "step in push" (move a step when self-attacking enemy is out of range). I abandoned it when I find something more effective. However, that is a feature that can be easily learned without target lines and actually learned by a lot of players. With target lines, there would be a lot of other features like "step in push" spreading in the leagues. So, even if there is no need to copy a large block of code except something so perfect like "Wozza's fire system", the spreading of features will make each other "similar enough", and thus retrains creativity. You should understand what I mean if you know why segregation is crucial to natural evolution.

Nevertheless, this is just my first point. Why do you ignore the second? Taking a step back, even if the lines do little harm, how much good do they do? Why shouldn't us follow Occam's Razor principle?

mcompany
Autonomous Entity
Autonomous Entity
Posts: 872

Re: Target Lines on Red Team

Post#27 » 01 Jul 2017, 19:05

And you are missin my point on accuracy. Recently, I purposely threw a bad AI on SYP specifically looking for how others could beat it and still somehow do so well against my AIs. And when I commented in telegram over how someone lost to it, y'all said that it wasn't that bad, not realizing that it was only a minor change to a functionally identical AI to the mission AIs. Please tell me how does one figure out that's what I'm doing (rather than make brand new rushing AIs). I believe the first usage of "push" to describe a highly aggressive medium range AI was describing my AIs, but my AIs never did that on purpose and that most came out of a design of having "move to the enemy" as the last bit of code and throwing a bunch of conditions to my actions to reach it. Yet how does one know that I don't have a node in the middle of my AI saying that I should push (and historically, I think that's what most people thought my AI had and tried to copy). And if I wrote 192 nodes to figure out who to attack next, how would you know that's what I did? How would you even know which version of the 192 nodes I did? This sort of information is just not possible to fully determine by battling it, with or without lines. In fact, the only real aid the lines are at this point is to avoid needing rulers and the such to do this same sort of work (stuff that I've been doing since forever). The lines don't just tell me whether or not you used a huge tree to decide that it was the correct target, and the lines don't tell you what order one decides their target. So no, it wouldn't matter if you are copying the whole AI, or a piece of the AI, seeing lines does not give you the advantage you need to copy Wozza's fire control system

Now, if you honestly believe "close enough" works, then go into test mode and play with you AI vs your AI with the attack logic being a bit different and tell me that your AI still works just fine. Unless there are several redundant nodes to catch edge cases where you almost do something you don't want to do, it is extremely hard, if not impossible, under any complex AI to get the same exact functionality out of slightly different code, or at least one of similar enough functionality. And what one could generate out of observation is most likely not going to be anything close to similar to the original.

Please actually copy some code first before saying that it is so much easier to copy code now that anyone could do it. I challenge you to actually stop having empty words and try it for yourself, and then come back to us.
Castlevania wrote:Nevertheless, this is just my first point. Why do you ignore the second? Taking a step back, even if the lines do little harm, how much good do they do? Why shouldn't us follow Occam's Razor principle?

Since when was this ever a second point of yours? Your original second point was that the line somehow points out the impossible (which simply isn't quite true), so where in the world you ever wrote this point?

Anyways, how does the lines help? How about that one moment on Mind Game against Ritter where I started complaining about how my AI reacted to Ritter's incorrectly when it turn out that Ritter bot was aiming at a different AI? How about debugging in general how action filters work? How about the ability to understand that you wasn't beat by magic, but because the enemy is better at focus fire than you, or just an ability to keep up with what is going on in general? I think a bigger question is that there aren't this super harmful thing with the lines, just simply a fear, so why is it such a necessity for you to get them removed from the game

pier4r
Skynet
Skynet
Posts: 3390

Re: Target Lines on Red Team

Post#28 » 01 Jul 2017, 20:50

I think mcompany likes the topic too much so he is building quite good arguments.

As I said, exposing "intentions" of the enemy takes away some difficulty of the game and I do not necessarily like it, but at the end given the complexity of the top ais, neither I think will be annoying. Let's say good players will catch up more quickly, and that is great.

Dominating just because the other cannot figure out what you do is nice, but also boring after a while.

I will dominate even after sharing all my ai, node for node, Muahahha.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

User avatar
Kanishka
Skynet
Skynet
Posts: 1421
Contact:

Re: Target Lines on Red Team

Post#29 » 02 Jul 2017, 04:58

Castlevania, there's no "Step In Push" ANYWHERE in my code. I just move when Enemy is Not In Range. :roll:
Fixes break an AI more than bugs do. :ugeek:

Gladiabots Off-Topic Chat


My Stats: Kanishka_RN3;Kanishka_MiPad

pier4r
Skynet
Skynet
Posts: 3390

Re: Target Lines on Red Team

Post#30 » 06 Jul 2017, 23:27

Surely debugging is easier (less time consuming) so I would expect improvements by strong players needing less games than before.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

Return to “Feature Requests”

Who is online

Users browsing this forum: No registered users and 1 guest