Currently Online

Latest Posts

Topic: Bechmark-play with Record-list

TuStudent
Avatar
Topic Opener
Joined: 2015-05-02, 23:30
Posts: 22
Ranking
Pry about Widelands
Posted at: 2019-03-06, 21:16

Is the a map where there is a benchmarkrecord like the most points after 4h for Collector (no opponent) or the fastest play to Autocrat on a specific map (with always the same opponent (no KI)).


Top Quote
WorldSavior
Avatar
Joined: 2016-10-15, 04:10
Posts: 2091
OS: Linux
Version: Recent tournament version
Ranking
One Elder of Players
Location: Germany
Posted at: 2019-03-06, 21:41

TuStudent wrote:

Is the a map where there is a benchmarkrecord like the most points after 4h for Collector (no opponent)

I don't know

or the fastest play to Autocrat on a specific map (with always the same opponent (no KI)).

How do you mean that exactly? Fastest autocrat wins against human opponents? I've defeated others several times in less than one hour, but I don't know exactly my record. However, notice that there are guys who play even weaker than the AI face-wink.png


Wanted to save the world, then I got widetracked

Top Quote
TuStudent
Avatar
Topic Opener
Joined: 2015-05-02, 23:30
Posts: 22
Ranking
Pry about Widelands
Posted at: 2019-03-06, 22:08

How do you mean that exactly? Fastest autocrat wins against human opponents? I've defeated others several times in less than one hour, but I don't know exactly my record. However, notice that there are guys who play even weaker than the AI face-wink.png

Fastest autocrat wins against one (or more) constant opponent (AI with solders and military building, but without any wares)

If you can win within one hour, you maybe have to make a stronger opponent.

I ment I do not want to have any relevant randomness (like different strong opponents). If someone uses the same strategey it should be 100% reproduciple, otherwise the benchmark would not be fair.


Top Quote
WorldSavior
Avatar
Joined: 2016-10-15, 04:10
Posts: 2091
OS: Linux
Version: Recent tournament version
Ranking
One Elder of Players
Location: Germany
Posted at: 2019-03-06, 22:58

TuStudent wrote:

How do you mean that exactly? Fastest autocrat wins against human opponents? I've defeated others several times in less than one hour, but I don't know exactly my record. However, notice that there are guys who play even weaker than the AI face-wink.png

Fastest autocrat wins against one (or more) constant opponent (AI with solders and military building, but without any wares)

The Headquarter is also a military building, so one could set the AI to "No AI" and it will do nothing, so the wares don't matter. Is that what you mean? Times depend then on map and tribe.

If you can win within one hour, you maybe have to make a stronger opponent.

You want to say that I should teach the opponents? face-wink.png

I ment I do not want to have any relevant randomness (like different strong opponents). If someone uses the same strategey it should be 100% reproduciple, otherwise the benchmark would not be fair.

A "No AI/ Idling Headquarter" excludes most randomness.

Edited: 2019-03-06, 22:59

Wanted to save the world, then I got widetracked

Top Quote
TuStudent
Avatar
Topic Opener
Joined: 2015-05-02, 23:30
Posts: 22
Ranking
Pry about Widelands
Posted at: 2019-03-06, 23:20

WorldSavior wrote:

TuStudent wrote:

Fastest autocrat wins against one (or more) constant opponent (AI with solders and military building, but without any wares)

The Headquarter is also a military building, so one could set the AI to "No AI" and it will do nothing, so the wares don't matter. Is that what you mean? Times depend then on map and tribe.

Yes exactly! There should be a Record-list for most Points (Collector) or shortest Time(Autocrat), for a representative map, splitted for different tribes and splitted for bots and humans. (Since it is "reproducible" benchmark this could be a way to improve "AI", maybe even stronger than the best humans.)

The idle "No AI" have to have more military buildings (with more soilders) to make the opponent stronger.

WorldSavior wrote:

TuStudent wrote: If you can win within one hour, you maybe have to make a stronger opponent.

You want to say that I should teach the opponents? face-wink.png

You should trade with the opponent (Whislist-feature) and send him wares and soiders. face-grin.png


Top Quote
WorldSavior
Avatar
Joined: 2016-10-15, 04:10
Posts: 2091
OS: Linux
Version: Recent tournament version
Ranking
One Elder of Players
Location: Germany
Posted at: 2019-03-06, 23:56

TuStudent wrote:

WorldSavior wrote:

TuStudent wrote:

Fastest autocrat wins against one (or more) constant opponent (AI with solders and military building, but without any wares)

The Headquarter is also a military building, so one could set the AI to "No AI" and it will do nothing, so the wares don't matter. Is that what you mean? Times depend then on map and tribe.

Yes exactly! There should be a Record-list for most Points (Collector) or shortest Time(Autocrat), for a representative map, splitted for different tribes and splitted for bots and humans.

Why splitted for bots and humans?

(Since it is "reproducible" benchmark this could be a way to improve "AI", maybe even stronger than the best humans.)

I think that AI experts would consider it as an incredible hard task to create an AI which plays Widelands better than the strongest players. Probably Widelands is much much more complex than Go ( isn't it?) and even in this game it was considered as impossible to create an AI which beats experts for a long time.

The idle "No AI" have to have more military buildings (with more soilders) to make the opponent stronger.

Just give the idle AI a trading outpost, this has more and stronger soldiers. And headquarters are hard to beat if they are full of soldiers, don't you think so? By the way Frisians have the strongest starting army. You could also design starting conditions with even more soldiers. But at the other hand combat tournaments are more interesting than a competition about who destroys a headquarter in the shortest time.

WorldSavior wrote:

TuStudent wrote: If you can win within one hour, you maybe have to make a stronger opponent.

You want to say that I should teach the opponents? face-wink.png

You should trade with the opponent (Whislist-feature) and send him wares and soiders. face-grin.png

It's already planned that trade will be a thing. But what would be the point of trading with the enemy? I think it would be only meaningful if one would train with the allies, which could be very interesting. By the way: Ports are good buildings for implementing the trading system.


Wanted to save the world, then I got widetracked

Top Quote
TuStudent
Avatar
Topic Opener
Joined: 2015-05-02, 23:30
Posts: 22
Ranking
Pry about Widelands
Posted at: 2019-03-07, 08:49

WorldSavior wrote:

Why splitted for bots and humans?

If this game is 100% repeatable, you know when something is happening. So you play it several times as a human read the inputdata and optimize the milliseconds in that why that way you get the best "deterministic randomnes" (like send two soildiers to a millitarybuilding, and both win the fight). I don't know how the randmones is calculated.

Therfore I assume it is possible to make a better AI, just for a specific map and just for a specific tribe. (The AI might fail completly if just one ware is different than expected.)

You should trade with the opponent (Whislist-feature) and send him wares and soiders. face-grin.png

It's already planned that trade will be a thing. But what would be the point of trading with the enemy? I think it would be only meaningful if one would train with the allies, which could be very interesting. By the way: Ports are good buildings for implementing the trading system.

It was a joke. (not serious)

Just because you questioned: But if you have 4 opponents, it could be an advantage to trade with somebody, when it could increase the probability of each of them to win.


Top Quote
WorldSavior
Avatar
Joined: 2016-10-15, 04:10
Posts: 2091
OS: Linux
Version: Recent tournament version
Ranking
One Elder of Players
Location: Germany
Posted at: 2019-03-08, 01:25

TuStudent wrote:

WorldSavior wrote:

Why splitted for bots and humans?

If this game is 100% repeatable, you know when something is happening. So you play it several times as a human read the inputdata and optimize the milliseconds in that why that way you get the best "deterministic randomnes" (like send two soildiers to a millitarybuilding, and both win the fight). I don't know how the randmones is calculated.

Does anybody know if an exact repetition of a match is possible or not (because of the random)? Both seems possible for me. By the way it seems to me that you didn't answer my question?

Therfore I assume it is possible to make a better AI, just for a specific map and just for a specific tribe. (The AI might fail completly if just one ware is different than expected.)

Only the start of the AI could be improved in this way, because later in the game it can be disturbed by it's opponents.

You should trade with the opponent (Whislist-feature) and send him wares and soiders. face-grin.png

It's already planned that trade will be a thing. But what would be the point of trading with the enemy? I think it would be only meaningful if one would train with the allies, which could be very interesting. By the way: Ports are good buildings for implementing the trading system.

It was a joke. (not serious)

Just because you questioned: But if you have 4 opponents, it could be an advantage to trade with somebody, when it could increase the probability of each of them to win.

Okay that's true, good point


Wanted to save the world, then I got widetracked

Top Quote
hessenfarmer
Avatar
Joined: 2014-12-11, 23:16
Posts: 2646
Ranking
One Elder of Players
Location: Bavaria
Posted at: 2019-03-08, 10:37

As far as I know the fights have remarkable randomness in them so I doubt that a game could be reproduced exactly. growth probability for trees is random as well as far as I know.

So for me a benchmark will not be reproducible. Although I admit this might be some kind of motivation to play widelands more often. Basically the last tournament was kind of a benchmark tournament.

Furthermore the AI is working completely different. It is (or should be) map and tribe agnostic.

Optimizing the AI in a way described makes no sense to me as it would require the human player doing exactly the things specified as well but this isn't realistic. In WL we have different tactics to hinder the opponent to easily conquer territory and this would never result in exactly one way to win a map.


Top Quote
GunChleoc
Avatar
Joined: 2013-10-07, 15:56
Posts: 3324
Ranking
One Elder of Players
Location: RenderedRect
Posted at: 2019-03-08, 19:27

Also, which tree and which bush gets planted is random. Mining also has randomness in it.


Busy indexing nil values

Top Quote