• Welcome to SC4 Devotion Forum Archives.

new traffic experiments

Started by ldog, October 23, 2009, 06:16:28 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

ldog

I do need to point out thought that most of the other sources I am reading they talk a lot about domain specific knowledge being more important than pure pathfinding.
Some of this is applicable to SC4 but most of it probably isn't. It is more relevant to pathfinding AI for wargames. Where one might want to give a bonus (seek and destroy) or a penalty (avoid combat) to things like enemy units along the path, or penalize open terrain (stick to cover).

Some of these bonuses and penalties in SC4 are clear to see, the travel preference for instance, and then the costs that go with them (or rather the costs are a penalty, the preference tells us when we should apply them). We give a penalty to using the non-preferred method of transportation. Which in effect is already throwing in "a bit of randomness" (yes, I know damn well there isn't a bloody thing that is random about any of it; it is simulating the appearance of randomness...so don't start telling me how wrong I am), one or the other would clearly be the best path but we are skewing it in favor of taking a path that isn't.

Intersection and turn capacity. Which by the way I just found a problem with them being linked and what seemed to be common design theory here.
If you raise the intersection capacity to accomodate the fact that 2 networks are there, hence giving a bonus instead of a penalty, you also do the same to turns. Whereas you certainly want turns to have a penalty (note: Steve scaled his back away from what was considered accepted...and now I understand why)

I am not sure what congestion vs speed should be considered. On the one hand the highest bonus we apply one could consider to be the base (1.3 x speed...such as for calculating PH, making it almost a part of the basic formula) or we could just look at it as a bonus/penalty system as well. Of course how we choose to look at it is not important; it is how the game engine looks at it.

ldog

#41
And back to our main man Patel again for a bit:

http://theory.stanford.edu/~amitp/GameProgramming/Heuristics.html
I am not going to quote this whole article because again it is long, it is all very good reading for anyone interested in the subject of A* but probably not so much for someone that just cares about SC4.

Of course I always enjoy a good chuckle:
"At the other extreme, if h(n) is very high relative to g(n), then only h(n) plays a role, and A* turns into BFS."
I love this guy.

I'm going to skip the rest of the section on heuristic for the moment. It has the usual formulas (which none of us seem to understand). I'll get back to it later.

Instead going to speed vs accuracy. This section is very interesting, a lot of it talks about things that once again don't really apply to SC4. For instance if you had plains and mountains but you don't want it to try quite so hard to go around the mountains so you actually use a heuristic that is not the lowest cost, but instead in between. We don't really have this kind of condition in SC4. UDI may be able to go offroading but traffic as we know it cannot. We don't want the engine to try to avoid using a particular kind of "terrain" (network in our case).

Some quotes:
"The choice between speed and accuracy does not have to be static. You can choose dynamically based on the CPU speed, the fraction of time going into pathfinding, the number of units on the map, the importance of the unit, the size of the group, the difficulty level, or any other factor. One way to make the tradeoff dynamic is to build a heuristic function that assumes the minimum cost to travel one grid space is 1 and then build a cost function that scales:

g'(n) = 1 + alpha * (g(n) - 1)

If alpha is 0, then the modified cost function will always be 1. At this setting, terrain costs are completely ignored, and A* works at the level of simple passable/unpassable grid spaces. If alpha is 1, then the original cost function will be used, and you get the full benefit of A*. You can set alpha anywhere in between."

Could be the variable we know as pathfinding heuristic is actually alpha. Would support Steve (Tropod, Mott) in that it is a constant no matter what the speeds we set in the exemplar are. If this is true, it isn't technically accurate to call it the pathfinding heuristic (it is actually a value in a formula to determine the pathfinding heuristic). I know also that it was Maxis in the internal documentation that called it that but as we all know the internal documentation does lie from time to time (and time and time again). Still want to know why my game calls it something else. Is it reader versions or is it USA vs international version?

"You should also consider switching from the heuristic returning the absolute minimum cost to returnin the expected minimum cost. For example, if most of your map is grasslands with a movement cost of 2 but some spaces on the map are roads with a movement cost of 1, then you might consider having the heuristic assume no roads, and return 2 * distance."

Now this. THIS is a lot more applicable to SC4. Replace the word grass with roads and the word roads (above in his quote not down here :P) with Monorail (or maybe even Subway/El or rail). That would support using the road (admissible...sometimes) heuristic instead of the faster (perfect) heuristic.

Last minute edit (and last peep out of me for the night, promise, closing the browser after this)
Forgot to mention that last page of reading also confirmed what I said above about the A* Manhattan Heuristic formula, 10 in the example above is indeed "D" which is minimum cost for moving from one space to an adjacent space aka minimum terrain entry cost. Just don't ask me why it is D instead of some other letter. Patel says it is D and that is good enough for me.
So to reiterate:
h(n) = D * (abs(n.x-goal.x) + abs(n.y-goal.y))

I also know some of this stuff was already quoted by Mott in discussions with Jason. Probably by Tropod and Steve as well. Excuse my laziness but it is a lot easier to reiterate it here than try to sift through old closed threads, which would probably result in me attempting thread necromancy on an unprecedented scale. Plus I have a much higher retention if I write (or type) it out than just reading it alone. So consider it indulging one of my (many) quirks. Thanks.

ldog

#42
So a funny thing happened today...j/k...I'll spare y'all

Today just going to keep it short.

A final point that needs to be added to the above (not for Steve or Jason but anyone else who may be interested and following along) of A* theory and how it relates to us.
Also this is not to say that the above presented material is complete nor necessarily accurate.

One of the general rules of A* is that it will ALWAYS find a path (so long as there actually is a path). Period. So how can what Steve and Jason are arguing about be true? How can the traffic engine cause abandonment in the first place?

Yes, it is true. Well then what about what I just said above? Also true. How can both statements be true?

In other applications of A*, like wargames in my above examples, the route will be found. Again, no ifs, ands or buts. A unit of course is likely to have a finite amount of movement points per turn so It also may take many turns to reach the destination.

In SC4 we don't have this luxury. We've got 1 turn (a single pass of the traffic simulator) to get to our destination and back. Network speeds are our terrain costs. Our "movement points" are our max commute time. From a practical standpoint in SC4 if one can't find a route that one can complete in 1 turn then one could say that we cannot find a valid route (even if not technically accurate).

I am sorry if this is so basic as to be elementary to everyone but me. As I did say in the beginning this is "traffic sim for noobs". Steve and Jason being our resident experts are the teachers here, even if I am the one asking the questions and being a downright unruly student.

z

Quote from: ldog on October 29, 2009, 05:54:57 PM
One of the general rules of A* is that it will ALWAYS find a path (so long as there actually is a path). Period. So how can what Steve and Jason are arguing about be true? How can the traffic engine cause abandonment in the first place?

Yes, it is true. Well then what about what I just said above? Also true. How can both statements be true?

A little bit of logic will tell you that there's exactly one way that both statements can be true:  SC4's implementation of A* isn't exactly what Amit is describing.  And that's actually the case.

ldog

Quote from: z on October 29, 2009, 06:14:08 PM
A little bit of logic will tell you that there's exactly one way that both statements can be true:  SC4's implementation of A* isn't exactly what Amit is describing.  And that's actually the case.

My little bit of logic was in the next paragraph. You are quoting me out of context.

I also explained how both statements are true but in a different way.

There go those words like "exactly one way" and "actually" again. Words that imply absolute certainty.
Unless you've seen the source code...
I disagree.The things that you have said could be exceptions, to paraphrase you "who knows what other kinds of calculations Maxis does in the executeable" , I have found relevant information about how these kinds of exceptions are valid and can be handled in A*. Although Amit may be the most read so far, he is not the only A* source I am reading.

So far nothing I've seen or that you guys have said during development of your traffic simulators proves that SC4 somehow breaks the rules of A*.

Now to read Jasons post.

ldog

Quote from: jplumbley on October 29, 2009, 06:48:06 PM
It is a theory, but to be honest I think this is how the game really deals with pathfinding, abandonment and how they two together act to effect other parts of the game.  It is probably by no means a "perfect" description of how it works, but a general idea and I am sure there are other things that are not taken into consideration that take this basic theory and make it more complex under the same basic prinicpal I have just described.

I didn't quote your whole post for brevity (lol me brief? :P )
I think it is a valid theory.

It also points out an error I made above; although that error doesn't necessarily invalidate what I said.
The algorithm CAN choose an "invalid" route. The algorithm does not know what is an "invalid" route or a valid route (no route is of course a seperate condition, such as we bulldoze all the roads, which is the 1 time when the algorithm can't find a route, any route). What makes the route valid or "invalid" would take place outside of the algorithm. Possibly in another part of the game. It could be what you suggest, it could be something as simple as "throw out any path not completed during a run" which probably doesn't even require much if any extra code. If they don't save the calculated but uncompleted route from run to run. It could also be done a few other ways (not going to enumerate them all), among them the way Steve says it is done.

Once again, none of it conflicts with A* theory. What the other parts of the game engine do with decisions the traffic simulator has made still don't change the A* algorithm, which we can take as a given that the game uses. The same as we can take for a given that it is 4-path (Manhattan) and not 8 or 16.

Looking back just around here and simtropolis again, I still haven't come across Tropod. I guess just a direct search for him would be more fruitful.
Rereading Mott's theory yet again, it still seems very sound (which is the basis of Z and A) but he stops short of answering how the perfect PH of .003 was calculated (Steve pointed out this was likely said by Tropod, and I believe him, I just haven't found it yet is all).

Also Mott himself alluded to having a very strong mathematics background. I don't know about you two, but I myself have no formal education beyond trig, geometry, algebra and calculations specific to electrical and electronic engineering that might be calculus (but are probably just trig) and some other calculation specific to computers and network engineering (except for some log functions probably none of that calculus either).

Now I can still puzzle out a lot of things that are beyond my education level, and of course I can punch numbers into a calculator and punch other buttons when directed to, but it is still going to take me some time to plot out some adjusted errors to work with. Right now that is what I am interested in because I am curious to see those values and where they stand in relation to .003 but if anyone has a linky handy I would also like to see how .003 was decided upon. Otherwise depending how busy I am tomorrow ( they actually expect me to do work at work, the nerve of some people  ??? ) I will hopefully find it.

I also agree, we don't need to understand the whole theory either. I am just trying to understand as much of it as I think is reasonably possible in a short period of time.
Ultimately if I go in and just start monkeying around as I have been doing; well this isn't work. If I make a bunch of bad decisions it isn't going to cause financial loss, breach of SLA, or even the kind of inter-departmental political debacles that seem to happen even when everything does go well...the worst that can happen is my computer will crash.

catty

Quote from: ldog on October 29, 2009, 08:58:29 PM
Looking back just around here and simtropolis again, I still haven't come across Tropod. I guess just a direct search for him would be more fruitful.
Rereading Mott's theory yet again, it still seems very sound (which is the basis of Z and A) but he stops short of answering how the perfect PH of .003 was calculated (Steve pointed out this was likely said by Tropod, and I believe him, I just haven't found it yet is all).

Try this topic  http://www.simtropolis.com/forum/messageview.cfm?catid=24&threadid=54310

Tropod says in it

QuotePathfinding Heuristic; 0.003000 (original value is actually 0.090000 - the lower the value, the more CPU intensive it is, but the pathfinding is more accurate)

:)
I meant," said Ipslore bitterly, "what is there in this world that truly makes living worthwhile?" DEATH thought about it. "CATS," he said eventually, "CATS ARE NICE.

ldog

Quote from: catty on October 29, 2009, 11:28:10 PM
Try this topic  http://www.simtropolis.com/forum/messageview.cfm?catid=24&threadid=54310

Tropod says in it

:)

Thanks Catty.
It looks like that puts me 1 step up and 2 steps back lol.
I still cannot find any post that says HOW .003 was determined.
Everything I read just references this as a given. Yet at some point someone in the community had to have presented how they came up with that number.
I am about ready to give up on the simtropolis forum.
Also even at one point it seems that it was Wouanagaine who introduced the pathfinding heuristic to Mott: http://sc4devotion.com/forums/index.php?topic=2665.0
So I begin to think the posts with the information I seek was lost to database troubles.

SC4BOY

Knowing Tropod, his approach was fairly thorough.. I expect he simply lowered it until it gave consistently good results and left it at that.. For him with his equipment at the time, he probably saw no improvement in the RESULT based on lowering it further.. Perhaps you might try the same experiment with your equipment. An experiment of this nature is likely a funtion of how much detail and time one wishes to put into the experiment and how sophisticated your tools for measuring the result is.. the same problem we have today.. The test has to be in the application, not in the math.

catty

Quote from: ldog on October 30, 2009, 09:40:56 AM
So I begin to think the posts with the information I seek was lost to database troubles.

An awful lot got lost last year when Simtropolis was hacked, have a read of this topic in the Omnibus (unfortunately the links its referring to I haven't been able to locate anywhere, but it does show the testing they were doing)

http://www.simtropolis.com/omnibus/index.cfm/Main.SimCity_4.Commute_Time_and_Pathfinding_Report

:)
I meant," said Ipslore bitterly, "what is there in this world that truly makes living worthwhile?" DEATH thought about it. "CATS," he said eventually, "CATS ARE NICE.

ldog

#50
Quote from: SC4BOY on October 30, 2009, 09:50:08 AM
Knowing Tropod, his approach was fairly thorough.. I expect he simply lowered it until it gave consistently good results and left it at that.. For him with his equipment at the time, he probably saw no improvement in the RESULT based on lowering it further.. Perhaps you might try the same experiment with your equipment. An experiment of this nature is likely a funtion of how much detail and time one wishes to put into the experiment and how sophisticated your tools for measuring the result is.. the same problem we have today.. The test has to be in the application, not in the math.


The thing is understanding the math can at least get us very close. So when ultimately we have to determine by experiment, we can cut down the amount of experimentation needed to obtain satisfactory results. I don't necessarily want to duplicate his (or Steve's) experiments. Especially if I can understand how such experiments were done and it is good enough for me then there is no need. I can do other experiments that assume the results of those experiments as "givens". I am not doing this just to be a pain in the ass to Steve or anyone else here in the community.

As some of the values I want to work with are not the perfect PH but instead derived from it, even if I didn't dispute that .003 was the perfect PH, I still need to know how the number was arrived at.

It is just my approach. In the end I may decide that there is no need for me to do a lot of the planned experimentation, or even to make my own traffic simulator. I may just find that Z or A or even B are the best compromise between one that does exactly what I want and what is actually attainable within the game engine. I will at least be a "well-educated consumer" . I want to give it my best effort though.

ldog

#51
Quote from: catty on October 30, 2009, 10:48:46 AM
An awful lot got lost last year when Simtropolis was hacked, have a read of this topic in the Omnibus (unfortunately the links its referring to I haven't been able to locate anywhere, but it does show the testing they were doing)

http://www.simtropolis.com/omnibus/index.cfm/Main.SimCity_4.Commute_Time_and_Pathfinding_Report

:)

Yeah, it seems a great wealth of information was lost. Not just about the traffic sim, but the whole game.

Thanks for that link. It's definitely interesting and hey! lots of pictures, a picture is worth a thousand words as "they" say.
Even if it doesn't provide the answer, I think when I piece it together with Mott's theory I may be able to figure out the other things I am looking for (see above reply to SC4BOY)
I also have other questions even if I am not asking them at the moment, and that thread also answers some of them.

ldog

So I had some time to sit down and read that omnibus article Catty linked me.
Wow, it's old. The information in it is pre-RH even. A lot of it is what was thought at the time but proved wrong (the charts especially, related to some of my very first questions about time).

The pictures and their accompanying text however are interesting.

For anyone who is just starting out trying to understand the traffic simulator but doesn't want to spend hours pouring over A* theory (or reading all my drivel) if you want a very quick visual on exactly what the effects of having a very high inadmissible pathfinding heuristic are, go look at those pics.

The stuff on return trip routes  ()what()
I ran some simple tests based on part 2 using owr.  :o
None of this has changed. I didn't even realize it, and probably wouldn't have without doing this simple test because in a normal city it is much harder to observe if you aren't specifically looking for it. The owr is kinda useless. Now I see why Steve and Jason don't even consider having to equalize it for NWM to be an issue.
Wish someone would have pointed this simple fact out to me lol.

ldog

Something else occurred to me. Is it possible some of the information I can't find is because it was discussed in NAM team private forums?
For example the earliest occurrence I can find of perfect PH == .003 it is quoted as "the NAM teams value"
Is there anyone from the NAM team who was around "way back then" still active that could shed some light on this?
I know you've obviously been on the NAM team a long time Jason but I am guessing this predates even you and Mott?

ldog

Quote from: jplumbley on October 30, 2009, 08:27:51 PM
Back in the early days of Tropod, Darkmatter, Buggi, the7trumpets, etc  Many discussions took place using MSN, and there is some discussions in a NAM Team thread at Simtropolis that is private.  In all honesty, that thread if it still exists does not get used anymore and I have never actually seen it, just know of it's existence.

There is no known way how 0.003 was originally concieved.  I would certainly quote the info for you if I could.

Yeah, I figured as much.
I will try a little more to figure it out from A* theory before I just move on, like I said to Steve, I don't consider the perfect PH as some kind of holy grail.
At some point I need to get the ball rolling.
Thanks again Jason.

Tarkus

I did have access to the aforementioned private thread at ST.  It's in a partially archived state and is near impossible to get to now, though I've managed to save a .pdf of it.  The earliest posts, however, date to March 2006, well after the "Perfect Pathfinding" simulator (Simulator E) had been developed, and there's absolutely no discussion at all of the traffic simulator.

-Alex

ldog

Ok so...I need to get started on laying out goals for my new traffic simulator. A planned project proposal and statement of work if you will.
A few posts ago I was seriously re-evaluating do I really even need to do this? Maybe I don't. Maybe some of my "hey, has anyone tried this? has anyone even thought of this? lets try this" actually has been thought about, and tried, and found to not work. So maybe one of the existing simulators while not ideal for me are the closest thing possible. I've devoted far too much time to learning the theory to not do anything with it. I've also in a way devoted far too much of Jason and Steve's time to my learning as well. So I also in some way owe it to them to at least try to do something useful with all this information. Even if all I find is that there really isn't anything new that can be done. No different ways to do things.

I also am not done learning theory to my satisfaction. Still some things to finish up, so consider this a very rough first draft.
***This is going to be very long, even for me. Also this is not some big "hooplah" about what is going to be the next latest and greatest traffic simulator. Not even close. Many of you may want to skip this. It is mostly for my own guidance, which I could just as easily put in a Word doc but I present here for anyone who may be interested.***

Now without further ado:

TRAFFIC SIMULATOR L
Why?
Why build yet another traffic simulator? Why not just use one of the existing ones?
Why not Z? I feel with its bias towards high commute and high population that it is geared for a playstyle that is not mine. What one could infer I think could make Z "better" would not make Z "better". It would make Z not Z. Z is well-suited for what it was designed to do. It is done to Steve's satisfaction and many other peoples satisfaction as well. It is done. As I am not a believer in "one-size fits all" it does not have to be for everyone. It does not have to be for me for me to still think well of it. In a nutshell even though I won't use it, if I had a friend just starting to play the game who was going to use the CAM and who wanted to build a large to scale region, and they asked me which sim do I pick? I would recommend Z.

Why not A or B? While I feel they are more suited for my playstyle, I do feel that there are some tweaks based off of Steve's research and development that could be made to them. Ok, so why not just leave it to Jason or better yet help Jason to tweak A since he has expressed continued interest? This possibility is not mutually exclusive to building my own. I can build my own and still give Jason whatever help I am able (and that he wants).

Also something that applys to all current simulators. They are all part of the NAM. As such they are bound by certain design restrictions. They must be compatible with the NAM. They must make compromises to allow some of the very creative things the other NAM team members are doing. So this brings up one of my other reasons.

To NAM or not to NAM
What are you saying Lenny, you don't like the NAM? :o Don't be silly. I do like the NAM, very much. I especially like the diagonal and orthogonal roads. The NAM adds a lot of much needed eye-candy. Some people might dispute it is eye-candy. It is all functional so how can you call it eye-candy? A lot of it is functional eye-candy. As the NAM proves, you can make it look like the road is going wherever you want it...however it doesn't change the fact that this is still Manhattan A* and we can only move in 4 directions. Now all that being said of course, don't think that I look down on eye-candy. Quite the contrary, I love eye-candy just as much as the next person. I love eye-candy so much I play with Adams very smooth slope mod, which makes me drop a string of F bombs everytime I try to lay a bridge and half the time I try to lay any other kind of network. Why do I suffer through that? Because I LOVE the way the smooth slopes look. I HATE the way the default slopes look. Does it change the way my traffic network works? Not one bit. In the end it is very much about eye-candy.

However, I do feel that because of that there are some possibilitys that might have been overlooked or that weren't overlooked but had to be abandoned for the NAM. We don't have the source code. There is only so much Maxis made accessible for us to change (and even then before other people made tools that accessibility was much less than today). We change what we can change and then after that it comes down to kludgy work arounds and tradeoffs.

Let me make one thing clear. I am not knocking anyones work. Any of you who has produced anything to share with the rest of the community deserves gratitude and respect for it, no matter how large or small the content was. You have shared with all of us at least 1 more choice for something we may want in our game and this community and others like it have kept this game alive 6 years after its release. Many of these "kludges" are the result of brilliant out of the box thinking and lots of time and effort on the creators part and I do not trivialize any of it. Still it all comes down to tradeoffs. We can't have everything. Some choices invalidate others (incompatibility). Sometimes our computer cannot handle all of the things we might like to add to the game. Sometimes there just isn't room for everything we might like in a 256x256 tile map.

That being said, my goal is not to be inNAMpatible. If I can be NAMpatible then all the better. The thing is by designing a traffic simulator completely removed from the NAM the only rules I have to follow are of the game itself. No holds barred. Once that is done then I will see what compromises I have to make to make it NAMpatible and then I have to pick and choose what is more important to me. Maybe I won't have to make any and I am making something out of nothing.

One of the initial reasons for this decision I have already pretty much invalidated already a couple posts ago.

Isn't it kind of selfish wanting to design a simulator just for yourself? Why should anyone in this community waste even a minute of their time with me?
No. It would be if I were unwilling to share it with everyone else. It would be selfish if I wanted someone else to spend a lot of time making it just for me. That is not my intention at all. I will share it with anyone who wants it. Further I will share my research with anyone who wants it.
Part of being a successful modder (successful anything really) is setting expectations.
I am setting my expectations. There are many different reasons people take up modding. To some people it is very important that their mod be appreciated and used by as many people as possible. Most people don't start out that way, but somewhere along the line for many people it starts to turn into something akin to a popularity contest, and noone likes to lose. Modding also tends to become much less fun then. When done solely for other people it becomes almost like a job. I won't bore y'all with the details but this is not the first game I've modded.
So foremost I do this for fun. Fun being relative of course, this is fun for a borderline obsessive-compulsive geek like me. Second because I see a need, no matter how small. It's a niche to fill. I am not looking to "compete" with the "big guys". Most of you are very happy with your current traffic simulator. Many of you probably think I am nuts just for being so obsessive about it. I have found over the years that not many people share the same view of fun that I do. So I am willing to fill the niche, even if I am the only one in it. Now of course being human like the rest of you, sure I would love it if other people thought I had something good and used it. There is no denying that. I just will not be disappointed when that number of other people turns out to be something like....oh...I dunno...3? Also if the only fruits of my research are to help someone else make something far better than I can, then even that is good enough for me. So no, I don't approach this with a selfish attitude. I just approach this with goals that set me up to be "successful".

Overall design objective
To achieve a fun, challenging, highly playable, realistic traffic simulator within the scale of the game. With emphasis on the single city as opposed to the region.

Design prioritys
It must not make the game less fun. Fun is the overriding principle. Challenge..well challenge is part of fun for me so maybe being redundant.
It must also not make the game less playable. So it must be "highly useable". Of course there isn't a lot of leeway in this part. A large part of the traffic simulators usefulness is transparant to the end user so....I probably need to sleep and finish this in the morning...it has to work well. Yeah, I am not making much sense at this point, least of all to myself. Time to wrap this up quickly for the night.
Realism? I enjoy a degree of realism, not at the expense of the other factors that make a game. If I loved realism so much I wouldn't be playing computer games. Still the idea of a simulator generally is to simulate something that is real. In as much as I can make it seem more realistic without making it less fun or useable (computer must run at an acceptable rate) realism should be strived for when possible.
Scale? There isn't anything we can do about city tile size. We can complain all we want that we can't build a realistic city in a 4km square, but in the end we must live with it. So we can cope in various ways. My way would be to make the 4km square seem larger.

Ok, I will finish for the night here, tomorrow to start drilling down into specifics.

ldog

Quote from: Tarkus on October 30, 2009, 09:43:47 PM
I did have access to the aforementioned private thread at ST.  It's in a partially archived state and is near impossible to get to now, though I've managed to save a .pdf of it.  The earliest posts, however, date to March 2006, well after the "Perfect Pathfinding" simulator (Simulator E) had been developed, and there's absolutely no discussion at all of the traffic simulator.

-Alex

Well thank you very much for looking Alex.
If nothing else at least I know I can stop searching there.
Happy Halloween!

SC4BOY

Quote from: ldog on October 30, 2009, 11:02:07 AM
The thing is understanding the math can at least get us very close. So when ultimately we have to determine by experiment, we can cut down the amount of experimentation needed

Welp, as they used to say when I was growing up, "When all's said and done, there's a lot more said than done."  I doubt the testing took more than 2 hours or so.. a binary search will zero in on a result very quickly I expect.. :)  As I mentioned I expect a quantitative tool is not readily available or even much topic of speculation; so you run it a bit, look at it.. and it's "hmmm... go or no-go?" So how do you decide? What's good enough?.. what increment +/- gives measurable difference? etc

ldog

#59
Quote from: SC4BOY on October 30, 2009, 11:02:48 PM
Welp, as they used to say when I was growing up, "When all's said and done, there's a lot more said than done."  I doubt the testing took more than 2 hours or so.. a binary search will zero in on a result very quickly I expect.. :)  As I mentioned I expect a quantitative tool is not readily available or even much topic of speculation; so you run it a bit, look at it.. and it's "hmmm... go or no-go?" So how do you decide? What's good enough?.. what increment +/- gives measurable difference? etc

I don't really understand what you are trying to say. Of course I just rolled out of bed and coffee is still brewing.

*some time later*

Ok, so you saying you are a man of action. Too much talking going on here for you :P
I enjoy theory just as much (maybe more) as application. As far as discussion though, theory is where we say a lot more than do. Application doesn't involve a lot of saying, it is really just about doing, but it is at the point where there isn't so much left to talk about.

2 hours, true/false find the answer very quick...No, absolutely not. Using that method we can find values that work well in the game and we can find values that work poorly in the game but we cannot determine the perfect pathfinding heuristic in that manner. It isn't a "feel good" value, it is a strictly defined value.

As far as what's good enough. For purposes of above discussion nothing I am going to find apparantly. The source code which we will never see.
For practical purposes? Like I said, I don't have to know. My OCD decided it was interesting for its own sake. I have done my due dilligence in trying to find out how others came to the conclusion and now I move on.

Or in other words; One does not need to use the perfect pathfinding heuristic; one simply needs to use a pathfinding heuristic that works