• 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.

z

Quote from: ldog on November 03, 2009, 04:59:32 PM
I still don't understand why I am seeing abandoned due to commute time when I have a highway and train system that has no congestion and the zones are far less than half a large tile apart. No Steve, I'm not using .003 for ph. Using .009...

There's your answer.  You might want to bump up your priority of checking .003.  Just a suggestion...  ;)

b22rian

Quote from: ldog on November 03, 2009, 04:59:32 PM
Yeah, we are all a bit obsessed. Considering that little gem my A* research pointed me to on SC4s site, Maxis had much less lofty goals.
This was the first SC that even had pathfinding in the traffic engine, so they were happy just to add it, forget perfecting it.
If we were actually getting a SC5 we might get better.
Although honestly I think what the community has done with SC4 over the years it probably is still more than we would have gotten out of an SC5.



Test mini-update for the day (last night's play actually). I didn't make a lot of progress. Was still noting anomolies. So I started a couple other tests (I had the sense not to slag the started one this time though). I still don't understand why I am seeing abandoned due to commute time when I have a highway and train system that has no congestion and the zones are far less than half a large tile apart. No Steve, I'm not using .003 for ph. Using .009 ...incidentaly 82x1.3=106.6 and 1/106.6~0.009 ... theoreticly if we find an open highway we stop looking for a better route. I still get a fair amount of train use as well. Even though I am still not convinced about .003 at some point I will switch it up and see if it makes a difference.




   Oh, there is no question in my mind were better off with what has transpired, in terms of traffic sim research.
First our pioneers , tropod and the 7 trumphets and later through the work and research of Mott, jason, and
of course especially steve.. From not just ability but on account of all their hard work and dedication we now
use some  pretty darn good traffic sims.. when compared to the original maxis traffic sim..  &apls

  from what i gather with what you have done so far.. i would say your main problem is having the PH set to
.009.. Although sure I encourage you to do as much testing on lower Ph values as you have time for.. i do
trust steve , tropod and others that .003 is a pretty good setting for PH.. but  try some other Ph settings
as well and see what you come up with .. keep in mind that by using this thread you have created to keep
us up to date on your testing results , not only furthers what your trying to accomplish but it quite helpful to
those interested in traffic,  in general.. as you know many of the areas in traffic are quite complex and theoretical so the more testing we can do, the better !!

thanks brian

RippleJet

Quote from: ldog on November 03, 2009, 04:59:32 PM
if there actually is no job available for those sims does it still report due to commute time?

Yes


Quote from: ldog on November 03, 2009, 04:59:32 PM
I have not seen any commercial or industrial abandonment yet, just residential, mostly R$ , some R$$ but no R$$$ (paying better attention this time).

Commercials and industrials would never become abandonded due to lack of workers.
They can actually keep operating even if nobody is working there... $%Grinno$%

Commercials may abandon due to low customers though,
which would happen if there's no traffic on the networks nearby.

ldog

Quote from: RippleJet on November 03, 2009, 10:42:25 PM
Yes

Well that is what I am inclined to believe at the time; that it isn't the pathfinder being stupid and not finding a job, the job just isn't there to find.
I am currently testing with .003 at the moment to see if it makes a difference.

Quote from: RippleJet on November 03, 2009, 10:42:25 PM
Commercials may abandon due to low customers though,
which would happen if there's no traffic on the networks nearby.

Which would be when it says "abandoned due to lack of demand" correct?

RippleJet

#104
Quote from: ldog on November 04, 2009, 07:22:18 AM
Which would be when it says "abandoned due to lack of demand" correct?

"Lack of demand" can be several reasons, e.g. if the demand is actually negative, but more often if the desirability of the lot has fallen below the abandonment threshold.
The number of customers as such is a factor in the desirability, but not the only one.

I cannot remember for sure if "lack of customers" alone can be a reason shown for commerical abandonment... %confuso


Correction:

Together with metasmurf (tack Andreas!) we checked all available reasons for abandonment.
There are only five of them in the game:


  • Abandoned due to lack of power
  • Abandoned due to lack of water
  • Abandoned due to low desirability
  • Abandoned due to commute time
  • Abandoned due to low demand

The third one is the one that you would get due to low customers in a commercial building, or due to any other desirability factor being too low (all in all, for any RCI type due to the desirability falling below the abandonment threshold).

The fourth one is the only one that you'd normally get in residentials, unless you cut off power or water.

The fifth one is the one you'd get if the overall demand for that RCI type is negative, and that can occur for all RCI types.

ldog

Thanks for that list Rip.

I'm still trying to get results with the different pathfinder, since as ...

Quote from: z on November 03, 2009, 05:21:14 PM
There's your answer.  You might want to bump up your priority of checking .003.  Just a suggestion...  ;)

Somehow, I just knew you were going to say that. I must be psychic (or is it just psychotic?  :-\ )
Then again we all knew that, you knew that too ;)

Still too early for me to tell. Wish you could swap on the fly and have the changes take place instantly since I'd like to examine the same particular point in time with the different engine, but rubbish in one hand and want in the other...

I am seeing more interesting routing though. Sims are taking the long way around some places, even though at a glance it would seem to not be a faster way; with no or minor congestion on the main routes. They are using every possible road to get where they are going. Still I am going to give the engine the benefit of the doubt that it really is a faster route, and even if it isn't well hey we wanted to see those people doing like Jason described how he goes home so I can chalk it up to that.

The thing is it REALLY makes no sense. The "better pathfinding heuristic" should still be overkill for the choices and availability presented. This would tend to prove what Steve and others have said, and prove to me that it really is "alpha" as I said earlier in the thread. Of course it is far too early for me to conclude anything; just speculating on my observations at this point.

As far as traffic simulator caused abandonment; if this is indeed what I am seeing and not just a case of there isn't actually a job for them to find, then I do agree with Steve that it is bullshit....errr...a bug. I know I said above I wasn't of a strong opinion about it. I am now. I really went overboard with the traffic network from how I would normally play in an attempt to get the test city to approximate what I expect it might look like after I had spent a long time playing it. The highway network is massive overkill, with rail thrown in for good measure. While I haven't sat down and counted out the tiles, there is no way they could run out of time without it being a bug.

But I am getting ahead of myself here...maybe like I said there just was no job to be found.

b22rian

Quote from: ldog on November 04, 2009, 03:01:00 PM

I'm still trying to get results with the different pathfinder, since as ...

Somehow, I just knew you were going to say that. I must be psychic (or is it just psychotic?  :-\ )
Then again we all knew that, you knew that too ;)

Still too early for me to tell. Wish you could swap on the fly and have the changes take place instantly since I'd like to examine the same particular point in time with the different engine, but rubbish in one hand and want in the other...




Think its great your Experimenting with different Ph's, Lenny  :thumbsup:

try to keep in  mind ,the larger your city , the bigger your population, and the more complex your transit
system becomes.. i think the more differences your going to notice using the different PH settings..
umm, im not sure you mentioned how large a city this test city is in terms of population ?

thanks for the update,

Brian

z

Quote from: ldog on November 04, 2009, 03:01:00 PM
Wish you could swap on the fly and have the changes take place instantly

You can certainly speed them up a lot.  Simply demolish all the abandoned buildings.  You should get new buildings growing fairly quickly.  This tends to be a lot faster than waiting for the abandoned buildings to become reoccupied, for reasons that RippleJet can explain far better than I.  You'll get new buildings whether or not the problem has been solved, since the old ones weren't abandoned due to lack of demand.  If the problem has been solved, the new ones will stick around; if it hasn't, they'll become abandoned, too.

QuoteI am seeing more interesting routing though. Sims are taking the long way around some places, even though at a glance it would seem to not be a faster way; with no or minor congestion on the main routes. They are using every possible road to get where they are going.

Yes, that sounds like perfect pathfinding in action to me.  The routes generated are much more complex than for higher values of the PH, which is what Jason said he likes.  The routes often do look surprising, especially if you have a complex traffic network, but if you do the actual calculations, you should find that they actually are the fastest.  By using so many different routes, the Sims avoid creating congestion as much as possible, as congestion significantly lengthens commute time.

QuoteThe thing is it REALLY makes no sense. The "better pathfinding heuristic" should still be overkill for the choices and availability presented.

My recent experiments are providing results to the contrary.  For example, I have been accumulating evidence (which I will present later) that the entire cause of the abandonment in Nate's city was that the PH was too high.  Since Nate's city was on a medium tile, all of his jobs were less than half a large tile away from his residences.  So when you said, "...the zones are far less than half a large tile apart," I thought, "Well, if he's even mentioning half a large tile as a job distance, there are going to be problems."  Throw in a complex traffic network ("I really went overboard with the traffic network from how I would normally play"), which causes the complexity of possible paths to shoot way up, and a value of .009 for the PH just isn't going to cut it.

QuoteBut I am getting ahead of myself here...maybe like I said there just was no job to be found.

Everything RippleJet has said here on this topic is correct, as usual.  However, if a city is built at all intelligently, it's actually rather difficult to run out of jobs.  It's still possible that you're short a few, although I think that's unlikely.  We'll soon see, won't we?

I have recently come across a little bit of evidence that the Perfect Pathfinding number may be slightly lower than .003.  It definitely isn't higher.  But it may be something like .0028, or even .0026.  Unfortunately, the city that provided that evidence is now long gone.  If you demolish all your abandoned buildings and any of the new ones become abandoned, you might try using a PH slightly lower than .003 and see if that helps.  You have an excellent test case here, and there's an opportunity to pin things down more exactly that rarely occurs.

ldog

Quote from: z on November 04, 2009, 04:09:24 PM
You can certainly speed them up a lot.  Simply demolish all the abandoned buildings.  You should get new buildings growing fairly quickly.  This tends to be a lot faster than waiting for the abandoned buildings to become reoccupied, for reasons that RippleJet can explain far better than I.  You'll get new buildings whether or not the problem has been solved, since the old ones weren't abandoned due to lack of demand.  If the problem has been solved, the new ones will stick around; if it hasn't, they'll become abandoned, too.

My recent experiments are providing results to the contrary.  For example, I have been accumulating evidence (which I will present later) that the entire cause of the abandonment in Nate's city was that the PH was too high.  Since Nate's city was on a medium tile, all of his jobs were less than half a large tile away from his residences.  So when you said, "...the zones are far less than half a large tile apart," I thought, "Well, if he's even mentioning half a large tile as a job distance, there are going to be problems."  Throw in a complex traffic network ("I really went overboard with the traffic network from how I would normally play"), which causes the complexity of possible paths to shoot way up, and a value of .009 for the PH just isn't going to cut it.

Everything RippleJet has said here on this topic is correct, as usual.  However, if a city is built at all intelligently, it's actually rather difficult to run out of jobs.  It's still possible that you're short a few, although I think that's unlikely.  We'll soon see, won't we?

I have recently come across a little bit of evidence that the Perfect Pathfinding number may be slightly lower than .003.  It definitely isn't higher.  But it may be something like .0028, or even .0026.  Unfortunately, the city that provided that evidence is now long gone.  If you demolish all your abandoned buildings and any of the new ones become abandoned, you might try using a PH slightly lower than .003 and see if that helps.  You have an excellent test case here, and there's an opportunity to pin things down more exactly that rarely occurs.

Well I don't have a tremendous amount of abandonment. I just think it is at a point where there shouldn't be any. Of course it is much more noticeable right now than my prior playing because I am specificaly on the lookout for just these kinds of things. The network isn't so much complex (it is certainly still too simple to fully test a traffic simulator as a whole) as it is way more than this city currently needs. I don't know about how intelligently it is built either, like I said it is very artificial compared to how I build when playing. Normally I would build out continuously from a core, not plop down multiple cores with an unsustainable (cost-wise) transit network such as it is. I think the highway network alone ran me half a million to build. I haven't even gotten it to a point where I feel it is reasonable to throw buses into the mix, let alone sub/el and mono. I don't really buy the not enough jobs thing either, although I have to do a little more investigating before I can dismiss it.

I'm sorry Brian I don't remember the pop offhand, it is still pretty low. I want to say 160k but I will go verify it tonight. Last night I put that city aside and I went back to playing a few medium tiles, per thy451's gamespot guide (which is pre-RH by the way), which was how I started learning to play. I find it useful to go back to for quick(er) tests, because I know what to expect from those city builds, so they give me a good baseline to compare to. Now I also got a bit of abandonment in "D'urberville" with .009 , I switched to .003 but it hasn't been enough years I think. I still had some abandonment and I think I even had some new abandonment since changing. I'm also pretty sure that there it isn't a case of not enough jobs. If it were R$$$ I was losing, then I would be inclined to think it could be, but it is R$ and R$$ so I am not buying it. I do have a bit of traffic (there are just roads and streets right now) but nothing even my little traffic simulator (which looks amazingly similar to B hard at the moment) can't handle. Moving on to "Aureliano" starting fresh with .003 I'm pretty sure I didn't have any abandonment there (but then I had a lot of commuters). Of course the game locked the computer up as usual around 1AM and I decided it was time to go to bed.

It is becoming more evident to me that we are dealing with a variable pathfinding heuristic calculation internally, which explains (to me anyway :P ) how we can have a perfect pathfinding heuristic that is independent of the "terrain entry cost". I would still love to know how Tropod figured that out originally, but I am about at the point where it is good enough for me. I will still consider the value we know as pathfinding heuristic to be "alpha" but that is really somantics at this point.

I think I'd go so far as to say at this point for someone just starting down the road I just started down a few weeks ago, if you are only interested in pathfinding as it applys to SC4 then don't even bother reading up on A* (I still found it highly interesting and worthwhile reading but that is me). All we can do is make educated guesses about what Maxis did internally. As Steve kinda said "it goes against what I understand about A* but this is what experimentation proved" (that was paraphrased not an actual quote) and Jason, who said he didn't really get into A*, but it certainly didn't stop him from making some of the biggest contributions to traffic simulation as we know it.

Now having said that, learning a bit of A* theory has helped me to be able to understand what I see going on enough to speculate on a lot of things. Like I said to SC4boy, it does help cut down on the needed experimentation time.
I want to point out that because of that theory some of the observations I've made so far are enough proof to me to accept certain things but I am not making any claims that what I have done so far in any way constitutes thorough testing. Or in other words I've proved a couple things good enough for my own satisfaction but I couldn't at this time provide proof to anyone else (except the morning commute time thing, that was pretty conclusive...again not anything someone else didn't say already either). It wouldn't be enough proof even for myself except that it corroborates a lot of what Steve, Jason and the others before them have said.

And the tests go on...

z

Quote from: ldog on November 04, 2009, 03:01:00 PM
I am seeing more interesting routing though. Sims are taking the long way around some places, even though at a glance it would seem to not be a faster way; with no or minor congestion on the main routes. They are using every possible road to get where they are going.

I should be a little more explicit in explaining what's happening here.  You know how Sims spread out on the RHW?  What you're seeing here is that same effect, writ large.  In other words, it's the speeding premium.  This premium makes Sims prefer routes that are just a little less busy than the ones they would otherwise take, even if there's no congestion.

I had first set the speeding premium in Simulator Z to be 10%, as such a premium is good for the pathfinder in that it helps provide tie breaking for A*.  But this wasn't high enough to get spreading on the RHW.  Simulators A and B were using 40%, and although that seemed high to me, I didn't have RHW in my older cities to test, so I just went with it, knowing it would work.  I did mention to people that I thought it was high, though, and it was later determined that 30% worked well enough, so that's what's currently used in Simulators A and Z.  I still think it should be lower if possible, at least for Simulator Z - 20% sounds about right to me - but testing would have to be done to make sure that this would still result on proper spreading on relatively short stretches of the RHW that were carrying light traffic.  And if 20% turned out to be too low, I'd want to try 25%.  If you want to do some experiments along these lines, feel free - they would be quite helpful.

ldog

#110
The CvS curve was one of those things foremost in my mind when I was referring to Jason's greatest contributions.
(not discounting all the above, it's just pretty clear and well explained so I don't have anything to add)
I have of course been playing with my own adjustments to it.

Now one thing that keeps sticking in my mind about the pathfinding heuristic, and why it doesn't behave the way one would think it should after reading a bit of A* that no one seems to bring up:
Everyone talks a bit about the effects of changes to terrain entry cost in terms of how much further around the pathfinder will go then to find another path, and the effects of the heuristic on that but not about when it shouldn't bother to go around.

For example my reasoning for using .009 as my initial ph had nothing to do with it being the generally accepted "better pathfinding heuristic".
It is the theoretical ph for an uncongested highway in the current simulator "L" .
To put the math into words, the instructions to the pathfinding engine in effect should basicly be "If you find an open highway then don't bother looking for a better route"

I was going somewhere with this.  %confuso Anyway. I still have some ideas but of course the abandonment thang throws a wrench into the thing.
Again, nothing Steve hasn't said.

So anyway, Gridlockshore pop 40k. This is happy news actually. It means I played for quite some amount of hours without hitting the save button (cause I am almost positive within a few percent that my last pop was much higher), so I should be able to get a closer approximation of what I had happening a few nights ago and see if I get the same abandonment issues with the ph of .003

Got to read all those old Maxis posts Cathy found for us. Really interesting stuff. I'm inclined to agree with Toroca. 7T rebuffed that by bringing up the listed minimum requirements, but I don't think Maxis tried very hard to find a suitable compromise for ph. As we've all seen, .09 gives us BFS.

z

Yes, at the the time that I came on the scene, it was generally believed that the PH was incapable of optimizing properly for speed.  But of course, it is.

Meanwhile, I was able to reconstruct my testbed where I thought there was some evidence that .003 was higher than Perfect Pathfinding, and I now have some evidence that this is true.  I had a few buildings that would repeatedly abandon at that PH, yet none of them do at .002.  I'll now try to narrow this down further.

ldog

Quote from: z on November 05, 2009, 03:17:59 PM
Yes, at the the time that I came on the scene, it was generally believed that the PH was incapable of optimizing properly for speed.  But of course, it is.

Meanwhile, I was able to reconstruct my testbed where I thought there was some evidence that .003 was higher than Perfect Pathfinding, and I now have some evidence that this is true.  I had a few buildings that would repeatedly abandon at that PH, yet none of them do at .002.  I'll now try to narrow this down further.

Very interesting

b22rian

Quote from: z on November 05, 2009, 03:17:59 PM
Yes, at the the time that I came on the scene, it was generally believed that the PH was incapable of optimizing properly for speed.  But of course, it is.

Meanwhile, I was able to reconstruct my testbed where I thought there was some evidence that .003 was higher than Perfect Pathfinding, and I now have some evidence that this is true.  I had a few buildings that would repeatedly abandon at that PH, yet none of them do at .002.  I'll now try to narrow this down further.

Well Steve,

                i wish I could help you out with testing this.. But using sim Z with its PH of .003,
is evidently just too good in my cities.. I scarcely can find any abandonment due to commute time
at all.. Despite the fact i have one city nearing 2 million population and 2 other cities over a million in
my region..  :P

z

Quote from: b22rian on November 05, 2009, 04:20:56 PM
But using sim Z with its PH of .003, is evidently just too good in my cities.. I scarcely can find any abandonment due to commute time
at all.. Despite the fact i have one city nearing 2 million population and 2 other cities over a million in
my region..  :P

That's why I love using the South Side of Chicago as a testbed - it truly is "the baddest part of town."  Many prototypes of Simulator Z died here, long before the world ever saw them.  So it also has the nickname "The Graveyard of Traffic Simulators."  If you can make it in the South Side of Chicago, you can make it anywhere.

Just ask Barack Obama.  ;D

Regarding the PH...
Quote from: ldog on November 05, 2009, 03:44:19 PM
Very interesting

But it gets even more interesting.  Basically, all I did was take the original release version of Simulator Z (v1.0) and run it on the Near South Side.  Sure enough, as I remembered, it had more problems with abandonment than the current version (1.2).  As I mentioned, these disappeared completely with a PH of .002.  But with the next value I tried (.0025), a very small amount of abandonment started coming back.

When I reran the current Simulator Z under the same conditions, I got the same amount of abandonment at a PH of .003 that the v1.0 version produced with a PH of .0025.

How do we explain all this?  Is the value for perfect pathfinding affected by other things in the simulator?  I don't think so.  In fact, I'll still stick with the value of .003 for the perfect pathfinding value until proven otherwise.  The definition of perfect pathfinding in A* is that all the paths found will be the fastest ones possible.  It doesn't guarantee that paths will be found, because standard A* assumes that paths will always be found (if they exist) for all values of the PH.  We have found that Maxis uses a modified version of A*, but we don't know all the details of how it's modified.  Clearly, they have put a limit on how much time the pathfinder will spend looking for paths - that's the only way you can get abandonment in a simulator like Z, where the maximum commute time is effectively unlimited.  But we have been assuming that the perfect pathfinding value is given unlimited time to find paths.  Although this would seem reasonable, we actually have no direct evidence that this is the case.

Why the difference in abandonment between versions 1.0 and 1.2 of Simulator Z?  The change happened in v1.2; I noticed it at the time.  And one of the main changes in v1.2 is that I adjusted various MT speeds downwards.  ::)  This made the simulator perform significantly better in a number of ways; reduced abandonment was just one factor.

So travel type speeds seem to affect abandonment; they may have a direct effect on what the perfect pathfinding heuristic is as well, or these two factors may not be connected.  It would take a lot more testing to see what's really going on here.  Based on my experiments, I could probably get rid of all abandonment in Simulator Z by lowering the PH to .0025.  But this would slow down the game noticeably, and as Brian's report shows, I'm not sure that this problem is seen often enough to warrant that.  It may also be that I can get rid of this last little bit of abandonment by adjusting the travel type speeds further, although I'm reluctant to do that as well without other reasons.  For now, I'll just leave everything as it is.

catty

Quote from: z on November 05, 2009, 11:09:38 PM
...  It may also be that I can get rid of this last little bit of abandonment by adjusting the travel type speeds further, although I'm reluctant to do that as well without other reasons.  For now, I'll just leave everything as it is.

I think it would be more realistic to have a little bit of abandonment in your city, in a five minute walk around my neighbourhood we have a partially abandoned factory, about a dozen houses and a shop that has only just be reopened as a restaurant after being empty and boarded up for the last five years

:)
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.

RippleJet

Quote from: catty on November 06, 2009, 12:21:22 AM
I think it would be more realistic to have a little bit of abandonment in your city

I'd agree with that! :thumbsup:
A simulator that doesn't lead to any abandonment isn't based on RL... ::)

z

Yes, that makes sense to me too.

It also means that I don't have to do any more work in this area.  ;D

Of course, there are other ways to get abandonment too, if you like to have a lot of it...  ;)

b22rian

Quote from: catty on November 06, 2009, 12:21:22 AM
I think it would be more realistic to have a little bit of abandonment in your city, in a five minute walk around my neighbourhood we have a partially abandoned factory, about a dozen houses and a shop that has only just be reopened as a restaurant after being empty and boarded up for the last five years

:)

  I agree here.. Im fine with a "little bit " of abandonment ... it gives me another little challenge and gives
me something else to work on.. Of course having to deal with it too much would be a pain  :thumbsdown:
So for me so far anyways.. PH of .003 looks like a pretty good setting..

  Quote from Z ..

That's why I love using the South Side of Chicago as a testbed - it truly is "the baddest part of town."  Many prototypes of Simulator Z died here, long before the world ever saw them.  So it also has the nickname "The Graveyard of Traffic Simulators."  If you can make it in the South Side of Chicago, you can make it anywhere.


    Yes actually I saw in another thread your chicago region was over 7.5 million.. my Florida region just
went over 5 million last night.. Thats a pretty big population difference 2.5 million.. i also recall from some pics
you have some amazing dense areas  in this region  :o
So its understandable you would have some abandonment problems in some spots..

Brian

ldog

Quote from: z on November 05, 2009, 11:09:38 PM
Regarding the PH...
But it gets even more interesting.  Basically, all I did was take the original release version of Simulator Z (v1.0) and run it on the Near South Side.  Sure enough, as I remembered, it had more problems with abandonment than the current version (1.2).  As I mentioned, these disappeared completely with a PH of .002.  But with the next value I tried (.0025), a very small amount of abandonment started coming back.

When I reran the current Simulator Z under the same conditions, I got the same amount of abandonment at a PH of .003 that the v1.0 version produced with a PH of .0025.

It makes sense. Even if we assume a variable ph formula, I would think there is still some range of values that would need to be adjusted according to terrain entry costs.
The thing about "Maxis using a modified version of A*" I highly doubt the algorithm they use is any different than any other A* implementation. It would take more math skills than I have to prove or disprove this of course. Speaking strictly of the pathfinding algorithm (as a distinct piece of the traffic simulator) A path will always be found if one exists, regardless of the ph, even if that path is to Quote Patel "BFS". The ph function is purely a matter of when to accept the path you've found or to keep looking for a better one. When all is said and done it really is a pretty simple thing.

Now what happens outside the algorithm is a different matter. Even though the traffic simulator is a pretty basic and limited thing, how it interacts with the rest of the game is where it gets complicated. Looking back on some of my earlier theory and conclusions, they are of course wrong. Steve has said as much. Of course being thickskulled..uhh...thorough..yeah, being the thorough person that I am I had to prove it wrong for myself. I still don't think the underlying A* mechanics of SC4 are all that different than any other A* application. The big difference is in the application itself. We aren't just trying to route a unit around the map, and then all the other interactions with said unit will take place within the absolute certainty that that unit is indeed in that particular spot on the map at the time. The traffic simulator is it's own little world here. So while I still find the theory valid, I was making incorrect comparisons between how A* interacts with a turn-based wargame and how it interacts with SC4.

So kinda like Newton in Steve's earlier example, we know and can see the effects of "nearest destination attractiveness" (this is what ph is called when I look at the exemplar) even if we don't truly understand it (because for all intents and purposes we know it's the ph, but it doesn't work the way ph is supposed to).

Now the key point where I am saying I was wrong is that the correct pathfinding heuristic value to use was not as narrow a range or as important as we were making it out to be.
Apparantly it is. While I still agree with the basic premise most of you are also stating, that some abandonment is tolerable. What my experiments showed me thus far is an "undocumented feature" aka "a bug". What I am seeing proves to me that Steve is correct about the abandonment being an issue. Within the parameters I set in "L" and the networks available there is no good reason why I should have had any residential abandonment. I really am not buying my own "no job actually available" defense. I really don't think I can prove conclusively one way or the other if there is or is not a job available either. Like Steve, I can prove lowering the ph stops the abandonment. The question still remains, why? Why indeed...

Quote from: z on November 05, 2009, 11:09:38 PM
Clearly, they have put a limit on how much time the pathfinder will spend looking for paths - that's the only way you can get abandonment in a simulator like Z, where the maximum commute time is effectively unlimited.  But we have been assuming that the perfect pathfinding value is given unlimited time to find paths.  Although this would seem reasonable, we actually have no direct evidence that this is the case.

This may help us explain why. It makes a lot of sense. I don't think any of us (you included) were assuming unlimited time, but we were assuming the time given within the max commute time at least.
So like I said above, given an acceptable path that can make the trip there and back within the limits of commute time, it seemed safe to assume that the trip shall be made. Not so, as we've seen.

Maybe there is some arbitrary limit somewhere in the executeable that we can't do anything about. Possibly it does have something to do with those values I mentioned in the developer exemplars. Maybe it is somewhere else, in some other exemplar. It's safe to say the max commute time doesn't determine it; or at least not directly in the way one would think it should.

I haven't been back to "Gridlockshore" to try it with .003 but I expect I will not see the abandonment; or at least see less. I so far have not seen any abandonment in any of my other concurrently running experiments, "D'urberville" started clearing up immediately, and while there is still a bit of abandonment left, I expect it to clear up in a couple years (didn't want to bulldoze as Steve suggested).

Currently I've started building another large tile "Gridlockcity" because one thing I did realize trying to use my shrunk down 3RR clone is that I am wasting far too much time "fighting the map". Between my perfectionist streak, the very strict slopemod that I choose to use, the roughness of the map...it is all far too much distraction (and we've all seen how well I handle distractions :P ). 
So I decided to go far simpler, full flat boring map, built some grids, dropped the rest of the map into the water and called it "good enough". I'll be able to hit a larger population faster, which is needed to test the things I had stated I was going to be testing in the first place.