Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Eagleon]

I think he meant support vector machines. It sounds like it's a type of binary or near-binary sorting system - I'm curious how they can be used in place of a GA solution, especially with multiple possible "correct" strategies.

[Fayrik]

It sounds like it's a type of binary or near-binary sorting system - I'm curious how they can be used in place of a GA solution, especially with multiple possible "correct" strategies.

Fuzzy Logic?

...Okay, sorry. That was a bit of a silly joke.

Also, this may have already been noticed, but using the latest build, "procgenrts.swf", every time it initializes, Team 1 has a random colour but Team 2 is always a black outline.. And it won't show any stats for Team 2 either. Team 1 never randomizes regardless of who wins or loses either.
It managed to get past that one time, and randomized both teams again, but all their stats where 0...
I'm guessing this must be something to do with the development/browser issue?
(I used to have the development version of Flash player, and in all honesty, I really miss it. I with Adobe would do away with all this downloads and plugins crap, and just have SWF files open in new windows.)

[Draco18s]

Also, this may have already been noticed, but using the latest build, "procgenrts.swf", every time it initializes, Team 1 has a random colour but Team 2 is always a black outline.. And it won't show any stats for Team 2 either. Team 1 never randomizes regardless of who wins or loses either.

Working as intended.  Team 2 units are all different from each other, so I didn't give them a little stat window at the bottom (as there'd be 12 of them!).  No-color was kind of a happy accident.  Because they're just randomized (or genetically set) the function with the random color never gets called, so I left it.  I could randomize their colors too, but as I've stopped exploring in this direction I don't see the need.

It managed to get past that one time, and randomized both teams again, but all their stats where 0...
I'm guessing this must be something to do with the development/browser issue?

Not sure on this one.  I did have a bug a couple days ago where the stats at the bottom weren't being update, but that should have been fixed.

(I used to have the development version of Flash player, and in all honesty, I really miss it. I with Adobe would do away with all this downloads and plugins crap, and just have SWF files open in new windows.)

But then you can't embed them for awesome sexysauce of flash navigation buttons (*puke*) and all that animated jazz!
That said, there are legitimate reasons to have embedded flash.  A lot of places use it poorly, but flash is a tool, just like anything else.

[Fayrik]

But then you can't embed them for awesome sexysauce of flash navigation buttons (*puke*) and all that animated jazz!
That said, there are legitimate reasons to have embedded flash.  A lot of places use it poorly, but flash is a tool, just like anything else.

I quite agree with the amount of vomit in this response...
But the trouble is, flash is trying to be everything. It can either be a first class script-based animation, video and games platform, or it can be an embedded navigation system that's totally over the top.
The trouble is, Adobe are right now, just proving that it can't be both.

Working as intended.  Team 2 units are all different from each other, so I didn't give them a little stat window at the bottom (as there'd be 12 of them!).  No-color was kind of a happy accident.  Because they're just randomized (or genetically set) the function with the random color never gets called, so I left it.  I could randomize their colors too, but as I've stopped exploring in this direction I don't see the need.

Aha, that would make sense.
But perhaps as a request, could I ask that team 2's large colour be set to something across the board (maybe, non randomized or only randomized at startup) then have the inner color change for each unit?
That way we can track what these units are more scientifically!

[Draco18s]

Aha, that would make sense.
But perhaps as a request, could I ask that team 2's large colour be set to something across the board (maybe, non randomized or only randomized at startup) then have the inner color change for each unit?
That way we can track what these units are more scientifically!

I would if I was continuing development in that direction, but I'm not.  When I fixed the logic issues and other things that cropped up, the program didn't find a solution anywhere near as quickly as it did originally (back BEFORE the GA stuff).  And when it did find one it was wrong.  I kept getting Rock Paper Mushroom.  Or Rock Paper Shotgun.  No matter what I did to insure that every unit that was pushed into the array was capable of beating the one before it turned out to be false at the end.  So 1 would beat 3, and 3 would beat 2, but 1 would also beat 2.

[Nivim]

 Which is why you want to try something other than genetic algorithms? Hm. Have you kept track of exactly what builds are rocks, papers, mushrooms, or shotguns? Such as, was the rock or paper in the RPM triangle the same as the shotgun in the RPS triangle? It sounds like it should just come out right just from tweaking the system's numbers.

Scaler vector machines, basically they can do what a ga does but more efficiently towards the local optima. We use them in sort search spaces cleanly a lot.

  I can see how the support vector machines could be used for this effectively (possibly scaler vector machines too), but the wikipedia article isn't going to explain how that will work, with Flash. *Uses Google.* There seem to be quite a few public lectures and texts on the subject.

[Draco18s]

Which is why you want to try something other than genetic algorithms? Hm. Have you kept track of exactly what builds are rocks, papers, mushrooms, or shotguns? Such as, was the rock or paper in the RPM triangle the same as the shotgun in the RPS triangle? It sounds like it should just come out right just from tweaking the system's numbers.

There are approximately 2300 different units that can come about.  None of them are inherently "better" AFAIK.
Just that when I'd get three that I could then go "1 & 2: FIGHT!" I'd get one that couldn't beat either of the other two (eg. a mushroom).  One of the other two, then, would be able to beat both of its counterparts (e.g. shotgun).  Its not like I can compare the two triangles and go "oh yeah, just swap these two."

Scaler vector machines, basically they can do what a ga does but more efficiently towards the local optima. We use them in sort search spaces cleanly a lot.

  I can see how the support vector machines could be used for this effectively (possibly scaler vector machines too), but the wikipedia article isn't going to explain how that will work, with Flash. *Uses Google.* There seem to be quite a few public lectures and texts on the subject.

I'll take a look, but I'll need something in Plain English before I'll understand it.  Anything with an "Abstract" up front with delineated "Introduction" (etc.) sections I'm not going to be able to read and comprehend.  I'm going to need a simple, practical, example and get walked through the conceptual steps before I'm going to be able to apply it.

[Nivim]

There are approximately 2300 different units that can come about.  None of them are inherently "better" AFAIK.
Just that when I'd get three that I could then go "1 & 2: FIGHT!" I'd get one that couldn't beat either of the other two (eg. a mushroom).  One of the other two, then, would be able to beat both of its counterparts (e.g. shotgun).  Its not like I can compare the two triangles and go "oh yeah, just swap these two."

I wasn't thinking about swapping them, but instead connecting the two triangles at that point and making a pyramid; just to see if it would work. It sounds like the statistics you've put together are producing more points to your shape than just three.

Scaler vector machines, basically they can do what a ga does but more efficiently towards the local optima. We use them in sort search spaces cleanly a lot.

  I can see how the support vector machines could be used for this effectively (possibly scaler vector machines too), but the wikipedia article isn't going to explain how that will work, with Flash. *Uses Google.* There seem to be quite a few public lectures and texts on the subject.

I'll take a look, but I'll need something in Plain English before I'll understand it.  Anything with an "Abstract" up front with delineated "Introduction" (etc.) sections I'm not going to be able to read and comprehend.  I'm going to need a simple, practical, example and get walked through the conceptual steps before I'm going to be able to apply it.

I added the word "tutorial" to my search, and got this! It seems hard, and I had to look up a definition, but I think it could do what you're after; figuring out what generated units go in what groups, and then sending the groups at each other.

[Draco18s]

There are approximately 2300 different units that can come about.  None of them are inherently "better" AFAIK.
Just that when I'd get three that I could then go "1 & 2: FIGHT!" I'd get one that couldn't beat either of the other two (eg. a mushroom).  One of the other two, then, would be able to beat both of its counterparts (e.g. shotgun).  Its not like I can compare the two triangles and go "oh yeah, just swap these two."

I wasn't thinking about swapping them, but instead connecting the two triangles at that point and making a pyramid; just to see if it would work. It sounds like the statistics you've put together are producing more points to your shape than just three.

It'd require doing the process multiple times in the same program session, which I'm not set up to do (having two different loops with the GA stuff kind of made it a mess, the third loop with the user specified tests really made it a mess).

[Virroken]

Posting to follow. Draco, you are too cool.

[Draco18s]

Posting to follow. Draco, you are too cool.

Thanks.

[Draco18s]

I added the word "tutorial" to my search, and got this! It seems hard, and I had to look up a definition, but I think it could do what you're after; figuring out what generated units go in what groups, and then sending the groups at each other.

Ack.  Too much higher order math!  I haven't done calculus in 5 years!  And I wasn't that good at it!
And is that linear algebra too!?  Ffffffffffffffffffffffffffffffff, I failed that class.

[Draco18s]

GOD DAMN IDIOTS.


They're everywhere, I swear.

Someone tried to tell me that what I'm doing is impossible because no one has ever done it before and "Lrn 2 Conform."

All while having no fucking clue how to program.  Instead he reduced the problem to "simple math" and came to a solution (rather than realizing that the solution was the output of the program rather than the program itself).

GOD DAMN -ING IDIOTS.

[Azkanan]

Umad?

[Draco18s]

Just a titch.  Guy asked me what I was doing and an hour and a half later finally realized I was talking about coding my own program (despite the fact that I said as much at the start of the conversation) and that he didn't know how.  Then asked me what language ("C++?") as if the swf I showed him up front hadn't been any indication.  All the while solving the problem I was writing a program to solve, completely oblivious to the fact that that wasn't the problem I was solving.

I got back to work on this today though and discovered an error that been plaguing me.  I'll take a look at my genetic version again briefly to see if it exists over there too and if that solves the issue it was having.