Currently Online

Latest Posts

Topic: Rebalancing the tribes?

GunChleoc
Avatar
Joined: 2013-10-07, 15:56
Posts: 3324
Ranking
One Elder of Players
Location: RenderedRect
Posted at: 2016-02-18, 12:14

I think these are good ideas to tweak the AI further - it is a bit off-topic for this thread though. Could you please open a new thread for it, or post it to an existing AI improvement thread?


Busy indexing nil values

Top Quote
Tibor

Joined: 2009-03-23, 23:24
Posts: 1377
Ranking
One Elder of Players
Location: Slovakia
Posted at: 2016-02-18, 12:57

Well I think weak point of AI is initial phase of building economy - first 1.5 - 2 hours, afterwards it I believe performs good enough. So I would not manipulate productions programs and so, but just add more starting wares - this would help to built the economy faster - providing there is a room for it and sooner overcome the problematic period.

Also it would be nice to have more feedback from users about AI, I mean average players... I even have an idea of automatic sending game results (when window with winner pops up) to our site, with some basic info, so we are able to have real-life feedback/statistics... But users dont like such functionality I know. But if there was a checkbox "I agree with ....." in that window with winner, perhaps users would not complain much.

There is another branch with AI waiting for merge, so perhaps next week it will be in trunk


Top Quote
king_of_nowhere
Avatar
Topic Opener
Joined: 2014-09-15, 18:35
Posts: 1668
Ranking
One Elder of Players
Posted at: 2016-02-18, 13:36

GunChleoc wrote:

I agree that this is tricky to test - maybe you could play against yourself in LAN mode and have a fight between the Barbarians and the other tribes to see how the soldiers fare with surviving. We could also hack the starting conditions to start a game with fully promoted soldiers?

We already know the surviving of soldiers by caculation of probability. By the way, I realized that I made a mistake in my previous calculations: adding +5 to the final barbarian attack would not be enough to kill a fully promoted soldier in 4 hits, unless extreme luck was involved. you need 224 point of damage to kill a fully promoted soldier, and current barbarian attack is 47-51. with the bonus, it would become 52-56, and the only way to kill in 4 would be to deal 56 damage every time. with a +9 on the attack killing in 4 would be certain, with an intermediate bonus it would be uncertain. I don't know if it is possible to specify "attack increases by 8.5 every level". Alternatively, attack increase of 9 per level (against 8 for other tribes) would work.

But anyway, we can calculate the exact probability of a soldier winning, whatever values we choose. that does not require playtesting. the issue would be how that compares with the cost of the soldier. we know the barbarian soldiers are more expensive to train, but are they expensive enough to compensate for their small advantage and for the stronger early game? That, I have no idea how to determine from playtesting.

Tibor wrote:

Well I think weak point of AI is initial phase of building economy - first 1.5 - 2 hours, afterwards it I believe performs good enough. So I would not manipulate productions programs and so, but just add more starting wares - this would help to built the economy faster - providing there is a room for it and sooner overcome the problematic period.

Also it would be nice to have more feedback from users about AI, I mean average players... I even have an idea of automatic sending game results (when window with winner pops up) to our site, with some basic info, so we are able to have real-life feedback/statistics... But users dont like such functionality I know. But if there was a checkbox "I agree with ....." in that window with winner, perhaps users would not complain much.

There is another branch with AI waiting for merge, so perhaps next week it will be in trunk

Depends on what you mean by "good enough". Because, with the same land, AI will still produce three to five times less than a good human. That's actually the whole point of the great escape solo challenge map, where the human will start building an economy two hours after the ai, will have much less land, and no access to coal forcing the use of charcola burners, but a strong human can still outproduce the ai with those limitations. enough to overtake its military before the limited resources run out. But I suppose you can consider that good enough, if ai performs worse at building an economy. I suppose one thing that could be improved is that the ai tends to leave too much empty space around; it could make the buildings closer.

As for getting feedback from users, I'm not sure how meaningful such a feedback would be. Keep in mind also that many people do not play until win condition is met, but only after there is no more resistance. After I killed the enemy army, I don't spend a couple hours conquering all the rest of the land; the game just isn't interesting anymore. And probably nobody plays on a losing game. if the ai is slowly conquering your land, you give up and stop playing, you don't reach the message window. I doubt we'd get a single report of a lost game.


Top Quote
GunChleoc
Avatar
Joined: 2013-10-07, 15:56
Posts: 3324
Ranking
One Elder of Players
Location: RenderedRect
Posted at: 2016-02-18, 17:05

king_of_nowhere wrote:

I don't know if it is possible to specify "attack increases by 8.5 every level". Alternatively, attack increase of 9 per level (against 8 for other tribes) would work.

Since 8 is actually 800 internally, it's not a problem to change it to 850.

But anyway, we can calculate the exact probability of a soldier winning, whatever values we choose. that does not require playtesting. the issue would be how that compares with the cost of the soldier. we know the barbarian soldiers are more expensive to train, but are they expensive enough to compensate for their small advantage and for the stronger early game? That, I have no idea how to determine from playtesting.

Neither do I - maybe it would be best to calculate as best as we can and release it with Build19, then we can tweak this again for Build 20 if necessary.

As for getting feedback from users, I'm not sure how meaningful such a feedback would be. Keep in mind also that many people do not play until win condition is met, but only after there is no more resistance. After I killed the enemy army, I don't spend a couple hours conquering all the rest of the land; the game just isn't interesting anymore. And probably nobody plays on a losing game. if the ai is slowly conquering your land, you give up and stop playing, you don't reach the message window. I doubt we'd get a single report of a lost game.

We would need to send statistics when the user leaves the game then, no matter if it has been finished or not. The consent message should be both very visible and very non-annoying at the same time though, and it should be opt-in.

Edited: 2016-02-18, 17:05

Busy indexing nil values

Top Quote
einstein13
Avatar
Joined: 2013-07-29, 00:01
Posts: 1118
Ranking
One Elder of Players
Location: Poland
Posted at: 2016-02-19, 00:56

I think that the message should be visible as default, but the player should be able to hide it in options. Something like prompt windows in web browser when closing multiple tabs (f.e.: in Firefox).

About rebalancing: now I trust king_of_nowhere, but as I remember, his calculations seems to be right face-smile.png


einstein13
calculations & maps packages: http://wuatek.no-ip.org/~rak/widelands/
backup website files: http://kartezjusz.ddns.net/upload/widelands/

Top Quote
GunChleoc
Avatar
Joined: 2013-10-07, 15:56
Posts: 3324
Ranking
One Elder of Players
Location: RenderedRect
Posted at: 2016-02-19, 09:48

Shall we change the attack increase to 850 then and merge the branch?


Busy indexing nil values

Top Quote
einstein13
Avatar
Joined: 2013-07-29, 00:01
Posts: 1118
Ranking
One Elder of Players
Location: Poland
Posted at: 2016-02-19, 12:15

GunChleoc wrote:

Shall we change the attack increase to 850 then and merge the branch?

I think yes.


einstein13
calculations & maps packages: http://wuatek.no-ip.org/~rak/widelands/
backup website files: http://kartezjusz.ddns.net/upload/widelands/

Top Quote
GunChleoc
Avatar
Joined: 2013-10-07, 15:56
Posts: 3324
Ranking
One Elder of Players
Location: RenderedRect
Posted at: 2016-02-20, 10:01

OK, done face-smile.png


Busy indexing nil values

Top Quote
einstein13
Avatar
Joined: 2013-07-29, 00:01
Posts: 1118
Ranking
One Elder of Players
Location: Poland
Posted at: 2016-02-21, 14:01

Finally I've created small project: random figths between soldiers.

Fights are between soldiers: level 0 (no upgrade), level 2 (only evade upgrade), level 7 (only iron upgrade), level 10 (full upgrade).

First result is between level 10 soldiers:

vs. bar_10 emp_10 atl_10
bar_10 54.3% 39.8% 35.4%
emp_10 66.8% 53.1% 48.9%
atl_10 70.6% 56.2% 53.1%

Fights are always began from one side. For example 70.6% from first column, last row means that alteantan level 10 soldier always began figth (versus barbarians level 10 soldier) and he win 70.6% of fights = experimental probability of win. I hope my calculations aren't wrong face-smile.png

Oh... Those above are "old" values for barbarians face-wink.png

EDITED: I've found a bug, so I've changed values

Edited: 2016-02-21, 14:58

einstein13
calculations & maps packages: http://wuatek.no-ip.org/~rak/widelands/
backup website files: http://kartezjusz.ddns.net/upload/widelands/

Top Quote
GunChleoc
Avatar
Joined: 2013-10-07, 15:56
Posts: 3324
Ranking
One Elder of Players
Location: RenderedRect
Posted at: 2016-02-21, 15:05

Thanks for the table face-smile.png

Can you calculate that for the new values as well, so that we can see how it is now?


Busy indexing nil values

Top Quote