Cityhouses distribution

Topic: Cityhouses distribution (Read 11789 times) previous topic - next topic

Cityhouses distribution

March 28, 2009, 01:04:14 pm

Okay, we do have a problem...

looking at the "spectrograms" (time & level) of city buildings, it's all it should not be. Ideally there should be strong, clean horizontal lines separated by darkness, but in reality it's fluff all the way from bottom to top.

To be addressed in future.

edit: Detached corrupted attachments. Can be seen at http://vs.simutrans.com/remote_images/spectrograms/index.html

Re: Cityhouses distribution

Reply #1 – March 28, 2009, 01:09:19 pm

I don't see any problem, the pictures are nice... Btw, do they mean something?

Ehem... What does time-level exactly measure? And why should they be cleaner?

Re: Cityhouses distribution

Reply #2 – March 28, 2009, 02:03:02 pm

It's the Matrix!

Ok, the cityhouses theory says that to get variety in cities, one should cluster them on a few levels, so that when advancing to next level, there are many alternatives to choose from. So on the level axis that translates to lines and nothing between them. Admittedly with a very large object base it could be instead achieved by spacing them evenly, which means that frequent increases in level would bring variety by being scattered in space (translate: leveling up happens in different places in different time). But past experience supports the first method more - I guess because then it depends only on human setting the data, while the second makes some ****umptions about how game treats growth.

What the charts show: time is time, level is level

It is a 2d chart, so you have two axes and the third is "upwards" - expressed by brightness. By graphing counts not only against level but also time, you get a good idea how the level situation changes with era. I think it is the best way to look at things, in the big picture. For example it is easy to see that after 1940 residental houses get a lot of variety in levels 10-20.

I am not really sure how to achieve closest ideal state. Just brute force requantization of level to, say, multiples of 5? Some histogram optimization could be more fitting for objects we have, but it should take into account changes in time. It would be nice to have also charts of how many houses on which level are used in games, depending on time...

To give it some clever names, what we have is a state of space of time ranges and levels, where are placed some objects. The task is to find in this space a state which fits best the criteria AND has minimal distance from original state. That is quite ambiguous - what is minimal distance, how much does it weigh against fitness?

Meh. For now it gives me some clue as to where these lv60 statues reported on German forum as taking over early cities should be introduced or their level changed.

Re: Cityhouses distribution

Reply #3 – March 28, 2009, 03:01:46 pm

How did you do the diagrams? Did you extract that from the dat files somehow? Or Excell?

Btw: Maybe you could add a quick sketch of how you think the diagram should look? It would help to understand.

Re: Cityhouses distribution

Reply #4 – March 28, 2009, 03:24:36 pm

Like this... lines.

I wrote a program, what else? I wouldn't put up with writing all the text myself

Pak128 repo has a load of useful thingies like this one.

I still have trouble interpreting these data by just eyeballing them, it needs some more work - probably a "thermometer" on side relating the different shades of green to numbers. And maybe using some colour map instead of just one colour.

Re: Cityhouses distribution

Reply #5 – March 28, 2009, 03:51:11 pm

Yes it would be good to know how many paks mean which shade of green.

Re: Cityhouses distribution

Reply #6 – March 28, 2009, 06:24:22 pm

My inbox is flooded by people asking for that tool! Order the complete pak128 maintainer's toolbox today and get discount:
http://forum.simutrans.com/index.php?topic=1805.0

Re: Cityhouses distribution

Reply #7 – April 01, 2009, 08:00:02 pm

Updated the tool again. Now it uses my prescciousssssss beloved JPL pseudocolour palette and gives some more info.

Look at the example charts for 128 and 64.

And many thanks to Sojo who found a bug! (that is fixed, too)

Re: Cityhouses distribution

Reply #8 – April 01, 2009, 10:39:02 pm

Interesting. Some conclusions I get from the graphs related with pak64:

There're more low and high level comercial buildings than the rest, and they're concentrated in 1960-2040 period
Residential building graphics is more distributed, although the most important period is 1920-1940.
There aren't any industry building until 1870, before this date people just relax and don't work...

Now with pak128:

There're a lot of medium level residential buildings (levels 10 up to 20) and most of them start from 1940.
Comercial buildings distribution by levels is more uniform, but mostly starts at 1960.
There're a lot of low level (0-5) industrial buildings, mainly concentrated in 1950-1970 period

We may conclude that long timeline experience is still pretty poor in both paks, but slightly better in pak64. We should review intro dates of all buildings to fill the gaps in some periods, specially from 1850 until 1920. Both paksets are highly concentrated in 1930-2040, and we should try to get at least from 1880 up to 2040.

I'm curious about industry. Could you do the same graphs for factories, VS?

Re: Cityhouses distribution

Reply #9 – April 02, 2009, 10:28:15 am

Another interesting find is that 64 has three times lower psg levels.

Generally it is easy to see how we got first the 1930-2000 covered and then reluctantly expanded.

As to factories, sure, that is easy, but I fear the kind of data is very different. Is there actually pakset with industry timeline?

Re: Cityhouses distribution

Reply #10 – April 02, 2009, 10:51:39 am

Well, at least in pak64 almost all factories have intro_year. You can sort them by intro_year & pax_level, for example.

Re: Cityhouses distribution

Reply #11 – April 02, 2009, 01:36:53 pm

Just some comments of pak64, which is not too easily seen from those graphs ...

For pak64 there is more or less a decent timeline. This graph does not tell the whoile truth, as the low levels in pak64 are mostly climate specific while the highrises later in the game are more uniformly found around the world.

Also before 1850 there are few large industries (apart from very few cities in the world). Most work was done in small houses (a carpenters house is not very different from a normal one, the most striking difference would be the blacksmith). Also I have a very limited number of buildings to choose from, thus I decided to skip industrie completely.

Also pak64 has a steadily increasing climate dependent top level of IND/COM/RES. Thus later cities will generate more p****engers, since with the industrial revolution travel became possible for everybody.