Discussion: optimum starting settings

[jamespetts]

I have started to do some play testing recently with Simutrans-Experimental 9.x (pre-release) and the pre-release of Pak128.Britain-Ex 0.7 so as to check that they are properly balanced, and also to chase out any bugs. One thing that I have been tinkering with in particular are the starting and map generation settings. I shall record my observations so far in no particular order. If anyone else has any useful observations to add to this discussion, I should be most interested.

I should preface these observations by noting that I tend to use very large maps (1024 x 2048 and above; my latest test map is 2688 x 1408, equivalent to 672 x 352km), so the observations are relevant to maps of that size (which, of course, take some time to generate, especially when using a debugger). I also tend to set a starting year of 1830: the beginning of serious passenger railways in the UK.

Firstly, it can be difficult sometimes to get a decent number of rivers, especially on the larger maps. I find that reducing the map roughness (4 or below) and increasing the range of permissible lengths to the maximum (i.e., minimum length: 1; maximum length: 4096) helps, too. If there are too few rivers, they never merge, and so one only ends up with the tiny streams going out to the sea, which is not very satisfying (nor useful, if one wants to use boats, although I appreciate that Pak128.Britain is somewhat incomplete on the boat side at present). With a decent number of long rivers, there'll be an interesting distribution of towns, as they will (with the new city generation features recently introduced) populate the river valleys even on otherwise inhospitable rugged inner terrain, creating interesting challenges of connecting the towns in different river valleys to each other.

Having a good mountain height (I set mine to 200 generally) gives an interesting terrain, and allows cities to cluster on lower ground quite effectively, which should hopefully allow for the formation, when all the towns grow, of larger urban areas of lots of towns all merging into one, while at the same time allowing for good long-distance travel over less populated terrain. I have yet to find the optimum "cities like water" setting; the default of 60 might be a tad strong, with most towns hugging the coast, so I am in the process of experimenting with a setting of 50 to get a slightly less predictable distribution (latest results: 50 seems to work well).

I tend to like cities to be connected to each other by road to start with, as this would have been how things were in the UK in the 1830s: I find that maximum road distance of 256 works fairly well, although a higher value might work better to connect things fully.

As to trees: these tend to have a disproportionately severe adverse impact on game performance (on my system, anyway), so the default values that come with Pak128.Britain-Ex are for quite a minimal quantity of heavily wooded areas, albeit with a light smattering of trees elsewhere.

I have been tinkering with the code for the forthcoming 9.0 release, and have adjusted the spread of the sizes of generated towns: I no longer get a large number of tiny towns - towns with only one or two buildings other than the town hall. I tend to select about 5 big towns, which seems about right for the UK in 1830. I don't tend to use the city clusters feature, as I don't think that it works very well yet - towns that don't fit in the clusters end up being tiny towns, and, if one makes a large number of clusters, or makes larger clusters, it doesn't seem to make much of a difference to having no clusters at all. That feature may need some work, I think. It should also be noted that, with the new city generation features, one may not get as many towns as one asks for if they won't all fit: check the chart in the city list to see how many towns that there actually are.

I am still working on alighting on an optimum passenger factor. In the current release of Pak128.Britain-Ex (0.6), the passenger factor is 14, which turns out to be far too low: in all the games that I have seen that people have uploaded, only a single-track mainline railway is ever justified between towns. This does not give a sufficient challenge in respect of adapting to capacity issues, nor provide enough revenue capacity in the early years. I am looking into a factor of 20 at present, which seems to work well in 1830 so far, but might be a little high for later years - I shall have to see. One of the things that makes a difference to the effect of the passenger factor in Experimental to Standard is the journey time tolerance feature: not all of the passengers who can get to their destination do, in fact, travel, requiring some compensation in the passenger factor. However, this needs to be linked with the median starting size of towns, which, for 1830, I am currently using at 4,500. This produces a number of quite sizeable towns: after the 5 "big towns", with populations of between 23,517 and 8,258, the population range is 8.088 to 264, with at least 1/3rd of towns over 2,000 and more than half over 1,000. With towns that large, I suspect that the passenger factor may be better off at 19 or even 18, but I shall have to investigate further.

I have yet to find an optimum number of industry chains for a map that size, although I suspect that 64-96 or so should suffice in 1830. I have yet to optimise the relative production values of the industries: as others have pointed out in previous discussions, there is much to be said for reducing the consumption of shops considerably and increasing their incidence by a like amount. I have so far been focussing on passenger transport, so any thoughts on optimising industry setup would be especially welcome.

Hopefully, however, this setup will provide the right balance of urban and sub-urban transport with long-distance transport by road, rail and water. Any thoughts on any of the issues discussed here are very welcome!

[dave62]

James,

I see that you suggest using a median starting size of towns of 4,500.  That being the case what would your recommended as a starting number of towns for a map size of, say, 1024 x 2048, given that figure?  Or are town size and distribution handled differently in version 9.0.

You speak about "the city clusters feature".  Again is this new or have I missed something?

Best wishes
Dave

[jamespetts]

Dave,

thank you for your reply: the distribution of city sizes and the distribution of cities on the map is indeed handled differently in the forthcoming 9.0 and later, although the same approximate range of city sizes should work: for 1024x2048, try somewhere between 300 - 500 cities (assuming that you have at least 3/4 land and no more than about 1/4 water).

[ӔO]

For factories, 64 - 96 is good. It would be even better if the main supply was located nearby the factory, but the final shop for the finished goods were located very far. Something like 24 - 32 tiles for grain farms to a grain mill, but 64 - 600 for bakery.

one thing that could be done to give less passengers in later years of the game is to lower the passenger factor for factories.

I've noticed that some factories give an amazing amount of local passengers, but it has to be very nearby another high intensity spot, like a city center, factory or an attraction to generate a sizable amount of passengers. For a inner city bus line, that's about 3 to 5 stops away.

[inkelyad]

towns that don't fit in the clusters end up being tiny towns

You have too many cities. Intended mode for clustered map was 'all cities inside clusters'.

Still it can be fixed -- just shuffle city_population (except first umgebung_t::number_of_clusters)
before calling stadt_t::random_place.

[jamespetts]

Inkelyad - the problem is that it's not possible to determine how many cities is "too many" for any given number of clusters in advance. And I'm not quite clear what you mean here by "shuffle city_population... before calling stadt_t::random_place" - what sort of shuffling did you envisage?

[inkelyad]

And I'm not quite clear what you mean here by "shuffle city_population... before calling stadt_t::random_place" - what sort of shuffling did you envisage?

See attached patch. But I will advise against it. It make clustered an not clustered maps too similar. Especially when number of cities is big.

You can debug city placement(see utils/dbg_weightmap.*) . Define DBG_WEIGHTMAP in compiler options and look for generated *.pgm files in simutrans directory. Any decent image viewer should display pgm format. Don't know how well it work under Windows.

[jamespetts]

So, without the patch we have the problem that cities outside clusters are too small, and there's no way of knowing the right number of cities to fit inside clusters in advance, and with the patch we have the problem that clusters have very little effect at all? Both strike me as somewhat unsatisfactory.

[inkelyad]

Why change anything? Right now it works more or less as intended -- it generate high-populated areas with less populated area between.

There is no way to predict 'how many cities too many', yes. But see attached patch. It will generate warnings about it after map generation. So player can adjust variables for the next map.

EDIT: I do same mistake again. 'welt' must not be used inside stadt_t::random_place. Patch file replaced.

[inkelyad]

James. My own experiments with the bigger median starting size show that site ~20000 will have big empty zone around even with default city_isolation_factor=1(assuming there is no water in said zone). Is it a problem?

[jamespetts]

By "site", do you mean "size" here? And how big is the empty zone? Thank you for the patch, incidentally - I haven't had a chance to try it yet, but, from your description, it seems like a sensible way of solving the issue.

[inkelyad]

For population ~20000 city itself will have ~11 tiles in diameter.
"Avoid zone" for it will have from 250 to 150 tiles. (depends on altitude).

Attached debug weightmap for flat 1024x1024 map. Left upper corner is lake. Big dark circle is 'avoid zone' for first city(population ~21000).

EDIT: never mind all this. I overlooked you put city_isolation_factor = 32 in simuconf.tab. So there is a way make 'avoid zone' smaller. But i think cityrules.tab is a better place.

[inkelyad]

EDIT: Sorry. Posted bug must be in separate topic

[waerth]

Interesting about the pax generation.

If I change the pax generation settings now will that influence my savefile? Or do I have to start from scratch?

(I was wondering why I had barely any pax going on my networks, while in other Simutrans versions and pax I have so much I cannot handle it usually).

Also can I change a setting so I get more intercity pax?

[inkelyad]

Also can I change a setting so I get more intercity pax?

I think it was discussed in Reload config
Open 'settings' window. Economy tab.

[jamespetts]

If you are using the latest version from the 9.x branch (having compiled it from source yourself), then, yes, you can change settings while the game is playing. You will be able to do this with the final 9.0 when it is released. Do not, however, try to do this with 8.2: there are serious bugs that will corrupt all the important values and make the game very difficult and unsatisfying to play.

As to the passenger numbers, I am in the process of investigating that in detail myself. The system for generating passengers has changed considerably in Experimental compared to Standard, and is now more subtle and sophisticated. I am currently experimenting with greatly reducing the proportion of "factory_pax" (from 33 to 15) and of "tourist_pax" (from 16 to , boosting the "passenger_factor" to 20 (from 14 in Pak128.Britain-Ex 0.6) and reducing the "passenger_routing_local_chance" from 40 to 37 and setting the "passenger_routing_midrange_chance" to 36. I shall report back with my findings as to what constitute optimum settings. If anyone else has anything to contribute in that regard, I should be most grateful :-)

[inkelyad]

Can we have dynamic max_*_tolerance and min_*_tolerance? I think people were more accustomed to long travel in 1700-1800.

[jamespetts]

This issue has been discussed before, and my view is that a static value is - ultimately - more realistic. People were only accustomed to travel taking longer in the days of old because travel, in fact, did take longer. In other words, those who did travel were those for whom the benefit of the journey outweighed the inevitable large cost in time of making the journey. Those for whom that was not the case simply did not travel. Consequently, fewer people travelled then than now. That people expected journeys to take longer did not alter that fundamental cost/benefit analysis, which underlies the basic economics of passenger travel. People were not more willing to travel for longer in the past: they just did not have the option to travel for less time, which they would have taken if they had it.

How long that journeys take in the Simutrans world depends entirely on factors actually simulated in the game (the quality of players' transport connexions, the quality of road connexions and the speed of private cars/horse carts), and it does not make sense, therefore, to approximate what can be simulated.

[ӔO]

I don't think there's a need for dynamic passenger settings, just change the amount of passenger generation in factories and city buildings.

The population boom and eventual mass immigration came around with industrialization.
By immigration, there were at least 2 types, one was where farmers moved into cities, and another was workers looking for jobs in countries that started industrializing.

So, to put that into simutrans, before 1870 or there abouts, farms required lots of workers, but after 1890 many people moved into cities were either commerce, construction or factory workers.

Workers needed in mines, quarries and similar didn't change much at that time, but what did change was who was working there.

One thing Sumatrans doesn't have are worker demands, passenger demands, for construction projects, but there is a contractor yard.

[inkelyad]

We need more local goods transportation then. Scaled down food chains, for example. One pub/bar/bakery/bucher in ~ 20000 city (about 3km walking distance in pak128.Britain in 1830) does not seem right for me.

[jamespetts]

Thank you for your thoughts. In the days before the industrial revolution, the farm workers all lived within walking distance of their jobs (and often on the farm itself), so it would not be right to have a very high passenger level for the early farms. The reason that people worked further away from where they lived was that they could do so because the transport was better: in other words, the journey times became within their personal tolerances.

As to scaled down food chains, that does indeed seem a sensible idea. I think that people have suggested before having shops with lower consumption values and a higher distribution weight. I'd be interested in thoughts on how to develop that idea - would the farms need changing, too, for example?

[ӔO]

scaled down industries?
how about a farm with grain mill that produces both grain and flour?
the grain is consumed by beer/ale brewery.

meat and fish, for sure, were a delicacy before industrialization and before refrigeration, everything was caught locally. Widespread consumption of fish, meat and fruits only started to catch on at around 1860~1880, after these perishables got cheaper due to faster vehicles.

the food variety before 1860 was limited to bread, vegetables, cheese and ale for commoners. Wealthy people obviously enjoyed a wider variety.

[jamespetts]

AEO,

interesting thoughts. I'm not so sure about the farms with in-built mills (did farms ever actually have this), but wind-mills are a good idea for industry. We just need somebody to draw graphics for them...

[ӔO]

I think I might have misinterpreted what you meant by scale.

for grain farm : mill : bakery, how about 2:1:8?
the factory chain calculator does make extra chains if it sees that there's excess being made at one of the production factories.

[jamespetts]

The word "scale" I had taken from Inkelyad - what had you thought that I had meant? However, those ratios look interesting. How would you imagine that this would fit in with the general reduction in the capacity of shops?

[ӔO]

by scale, I originally thought you meant how long the chain was.
Now I see you mean the total amount of consumption and production.

2:1:8 was a ratio from a medieval game's bakery chain.
I give this ratio as a possibility, but if wiki is right, there were some 10,000 wind mills by the 19th century in Britain, so it seems like there weren't that many compared to bakeries and farms.

For the shops, I think there would be many local shops serving smaller areas, because people didn't travel as far and this should include buying groceries as well.

thinking about it more, maybe it should be 6:1:24 or there abouts.

[jamespetts]

Hmm, interesting: one mill serving 24 shops (and therefore, probably, 24 towns); is that number a guess or based on research? That seems like a far higher ratio than is normally seen in Simutrans, and I'm wondering whether setting it that high might have unintended consequences...

[ӔO]

my numbers are not based on any research and are purely speculative for game purposes.


some research seems to indicate that it takes 0.5ha of grain corn farm land per 2 people consuming 5000kcal/day without aid of farming equipment. With a tile size of 250m^2, or 6.25ha, one tile of farm land will support 25 farmers.
http://www.culturechange.org/cms/content/view/585/66/

[AvG]

I am experimenting with Exp8.2+Brit0.6 starting in 1750. Bits_per _month 21
- Adapted all factories to output less then 100. The slow speed and low cap of land vehicles make it IMHO to difficult to have a sufficient transport-system for large industries.
The max-number of vehicles on a road is restricting the manageable output of factories. If you do not fullfill in transporting their output they turn red and output becomes 0 for a long time; also the stock in the loading bays is gone. If this happens in a later stage of the factory-chain the refund system might do a lot of harm.
I aim for: Lots of small commodity-places (like farm, mine), few big real factories (like mill, slaughter-house) and again lots of small end-users (like pub, butchery)
- Using passenger-factor of 8. Having values of 16 or 20 looks nice, but again you have no means of transporting lots of people.
- Using high towngrowth-factors of 1150, 1600 and 2000, to prevent towns to grow to fast. However, they still grow to fast probably due to my action in them.
I have to quit my running Italy-scenario now because of very large negative revenues. All roadstops are green, all lines make a profit and still those **** negative revenues.
These started I think in the 3rd year and are becoming larger and larger. At present I can make a profit of ~20K/month. The latest negative influence was ~50K.
If I go on I will go bancrupt.
Refund system?? Bug??
AvG
Edit: Sorry for the ****. The expression looked to me quite harmless.

[inkelyad]

- Using passenger-factor of 8. Having values of 16 or 20 looks nice, but again you have no means of transporting lots of people.

Hm. People don't like spend whole week in travel. But they are usually fine with 'go to big town every monday'. It should be simulated somehow.

- Using high towngrowth-factors of 1150, 1600 and 2000, to prevent towns to grow to fast. However, they still grow to fast probably due to my action in them.

If towns size (not population) is a problem, you can change renovation_percentage in cityrule.tab Current 38 to 80, for example.

[AvG]

I think that passenger-factor 8 i.s.o 16 only means halving the amount of POSSIBLE passengers, before checks are made. That means that a POSSIBLE passenger can decide NOT to travel because the yourney will be to long.

Might be a good idea. At present my experience in 1750 is that town-size is not only growing to fast in population, but also in tiles. In other words: the density could be higher. However, I don't have any experience on that matter.
Your example, changing from 38 to 80. Is that a realistic proposal? What is the effect?
AvG

Edit: The renovation_percentage seems to be only 12% as standard. I did not touch that. My next start will use 25%

[AvG]

James,
I have found an old map showing in 1820 traveling-times via road and water.
After some math the conclusions:
- Average travel-speed ~ 4 km/h by road
- Average travel-speed ~ 4 km/h by waterway.
So the travel-time via waterway and road should not differ to much.
Also is visible on that map that lots of roads in these days were simple sand-roads. (In Simutrans-terms less than dirt-roads)
To simulate this in a proper way I would suggest to use for city-connection-roads the type dirt-road (when starting in 1820.

Sailing vessels.
In my 1750 scenario I used for the first time sailing vessels; the Norfolk Wherry, the brig and the dogger.
I notice some behaviours:
-The Norfolk Wherry is always at 4 km/h
-The Brig is doing 15 km/h when empty and 4 km/h loaded.
-The Dogger is allways running at max speed of 12 km/h

My first question is: What kind of speed-behavior is meant ?
Tweaking the proportion of power versus weight makes it possible to get the situation that every ship runs when empty at max speed and ALL 4 km/h when loaded.
But again: What do we want?
AvG

[inkelyad]

That means that a POSSIBLE passenger can decide NOT to travel because the yourney will be to long.

I think in current implementation 'wait 7 days (at home) for fast train' will be counted as journey time. => passenger will decide not to travel.

Your example, changing from 38 to 80. Is that a realistic proposal? What is the effect?

The density will be higher. Don't know how realistic it is. I don't think that simutrans city model is enough to simulate real-life cities.

But there is no hight level buildings in 1750. So effect can be small.

[neroden]

I'm back....

James, I think you have too many cities.  There's no "wilderness" on your maps, is there?

I've been running maps of similar size (1024x2048) with 64 cities; admittedly I often have around 50% water, but adjusting, 100 cities would be appropriate for your level of water.  300 cities?  Well, my cities are *also* probably twice as large as yours, but still.

As to the passenger numbers, I am in the process of investigating that in detail myself. The system for generating passengers has changed considerably in Experimental compared to Standard, and is now more subtle and sophisticated. I am currently experimenting with greatly reducing the proportion of "factory_pax" (from 33 to 15) and of "tourist_pax" (from 16 to , boosting the "passenger_factor" to 20 (from 14 in Pak128.Britain-Ex 0.6) and reducing the "passenger_routing_local_chance" from 40 to 37 and setting the "passenger_routing_midrange_chance" to 36. I shall report back with my findings as to what constitute optimum settings. If anyone else has anything to contribute in that regard, I should be most grateful :-)

This seems like the correct direction to adjust things in.  You might have cut factory and tourist pax down too much, but it's the correct direction to adjust them in.  It's also clear that we needed fewer short-range trips and more medium-to-long-range ones.

The more important problem in the early game is factory *goods*, where the entire balancing is broken, but this is a pak question mostly.

[inkelyad]

It's also clear that we needed fewer short-range trips and more medium-to-long-range ones.

I don't think so. Passenger traffic must not be too profitable. But industry input/output should depends on it. Can we calculate shop consumption from  traffic to it?

[jamespetts]

Welcome back! As to the number of cities (more accurately, conurbations), I am trying to simulate England, where, in certain parts, towns and villages can be very dense - have you ever been to South-East England and the home counties, for instance? If you have a look at a Google map of the area, you'll know what I mean. There need to be large areas as dense as that, and other areas rather sparser.

As to local versus mid-range and longer distance passengers: in the early game, it is the mid-range passengers that provide most of the profit, since local passengers generate only low levels of revenue, and the networks are insufficiently developed to transport long-distance passengers. Whilst I have been very busy over the last few weeks and have not had a chance to run the tests that I hope to run, I suspect that the issue is, in part, that fewer local passengers exceed the journey time tolerance, thus exaggerating the bias in favour of local passengers created deliberately (as this is realistic) to excessive levels.

What is the problem with the goods in Pak128.Britain-Ex, incidentally?

[ӔO]

I can tell you that with passengers = 16, there's more than enough passengers generated.

the key to success seems to be frequent service with faster connections at spots that will generate a lot of passengers.
I'm not even done building up the network, but it's earning a lot of money.
It just requires a large initial investment.

it seems to be better to not make tram hubs, because it messes up the spacing, and therefore the frequency of the service on the line.

run the vehicles and get them spaced out evenly, before you build the stops, or you will lose passengers that are waiting for the return of the first convoy in that line.
-----------------------------------------------------------
save file updated.
http://hotfile.com/dl/85368199/91ead1f/Le_monde.sve.html

[neroden]

Welcome back! As to the number of cities (more accurately, conurbations), I am trying to simulate England, where, in certain parts, towns and villages can be very dense - have you ever been to South-East England and the home counties, for instance? If you have a look at a Google map of the area, you'll know what I mean.

Subtle point here: I'm starting in 1830!  It wasn't that dense in 1830!  The smallest "villages" shouldn't even appear on the Simutrans scale, to start with.  If I'm doing my job, the city growth should cause that area to become extremely dense later (and it does, it does), but it should build up *around* my railroad lines.  :-)  (Eventually we will figure out how to have the game automatically generate "new towns" around appropriate levels of factory traffic. :-) )

I suspect that the issue is, in part, that fewer local passengers exceed the journey time tolerance, thus exaggerating the bias in favour of local passengers created deliberately (as this is realistic) to excessive levels.

I'm certain this is what was happening, and it seems to have been fixed now.

What is the problem with the goods in Pak128.Britain-Ex, incidentally?

The core problem with farms and mines producing too little to ever handle profitably has now been fixed.

The secondary problem, unfortunately, is that the income from 0% speedbonus and 3% speedbonus goods is simply too low to pay for the infrastructure of a railroad, or even a full-length road.  The volumes are high enough to overcrowd the roads but barely break even using horse-based road vehicles -- it's not quite profitable enough to justify building the roads.  This is exacerbated because many of the factory chains contain ONLY 0% and 3% speedbonus goods. 

My baseline is that hauling coal to a coal merchant with a steam engine and the cheapest bulk wagons, and the cheapest rails which can carry that steam engine, should be profitable, which it was at the time.  It's not even profitable to run it with a plateway and horse-drawn vehicles at the moment.  The running costs and infrastructure maintenance costs for all this stuff is very low already, so I think the base price of bulk goods needs to be tweaked upwards a bit (not much, maybe from 53 to 54).  As for the 3% speedbonus goods, either the base price needs to be raised, or the speedbonus percentage needs to be increased -- you have to have a *huge* (completely unattainable in 1830) difference in speed to get any price difference for these goods, and I think that's not what's intended.  If this is true across the timeline, perhaps these should be 4% or 5% speedbonus goods instead.  If this is only true in the early timeline, the speedbonus structure for the early timeline needs to be reexamined to figure out why this is happening.

[Václav]

I often start with city size of 1300 and with 13, 21 or 31 cities (based on map size; but I often play on map of Czech republic with size 2000*1316) - and with industry 4.

It gives game right difficulty. So it is not too difficult - but still it is not easy. With these settings there in game one of two very big city where you can start with passenger transportation. And there is also optimum density of industry - one or two chains ending by any shop what are quite distant from each other - and some powerstations.

If this is what you ask, James.

[ӔO]

a few minor bits I'd like to see.

pedes_and_car_info: off
tree_info: off
ground_info: off
townhall_info: on
only_single_info: off