Computer player refuses to mine gold

[GoogleCodeExporter]

Start the attached testmap in r1813. Notice blue refuses to mine gold at all 
and will go broke even though his base is surrounded by gold.

Behavior is related somehow to map layout. This map behaves the same in 0.4.5.

Original issue reported on code.google.com by Loobinex on 4 May 2015 at 5:52

Attachments:

• pleasemine.7z

[GoogleCodeExporter]

Additional note: The CP doesn't seem to mind not having gold at all, he will 
not assign creatures to the workshop either.

I've tried putting a portal on the map to see if this would make a difference, 
but it did not. Blue would dig for the portal but not go after gold.
Looking at compuchat the CP did continue to try and pick up gold from the floor 
after all the gold on the floor was gone.

Original comment by Loobinex on 4 May 2015 at 10:26

[GoogleCodeExporter]

The problem seems to be caused somehow by the fact that the dungeon is built 
directly against the gold. If some dirt is placed between the gold and blue 
dungeon the gold will be mined.

Original comment by Loobinex on 4 May 2015 at 11:40

[GoogleCodeExporter]

Have observed behavior like "not digging for gold" in campaign. Will 
investigate when I have time as a goal of mine is to make AI be a decent 
challenge - bugs like this completely destroys reason for me to play campaign 
since there's no opposition. :(

Original comment by eatpajdi...@gmail.com on 11 May 2015 at 9:17

[GoogleCodeExporter]

Getting a more challenging CP would be very nice. Having the CP not mine gold 
goes beyond ‘not challenging’, it is simply broken. Either you win straight 
away without a fight, but it is also possible that a map becomes unwinnable. 
For example for issue #612 I received a message from somebody who was replaying 
the NG+ campaign and found out that in level 5 the CP walled himself in. So 
fixing bugs with the CP is very important, but I say a different matter 
entirely.

For actually making the CP more challenging it needs to adopt some simple human 
like behaviors. Also note that there quite a few limitations a mapmaker can 
work around by giving the CP a pre-build dungeon, a different creature pool, 
fewer spells to research or even starting creatures, but also limitations that 
currently have no workaround. So keep in mind when making improvements to the 
CP if you are going for challenging custom maps, or if you want a challenging 
opponent in -1player mode.
I think currently the biggest gameplay improvements to the CP would be:
-   Better focus on training.
-   Research only with creatures that are good at it (if available), and further 
reduced(near zero) focus on research when all rooms are researched.
-   Getting a strong creature composition by getting rid of weak creatures.
-   Building treasure rooms close to gold.
-   Building libraries with a single entrance or out of the way.
-   (Re)build and defend bridges to attack, mine, collect prisoners etc. (Right 
now capturing and selling his bridge makes you safe from an attacking keeper.)
-   Capture and converting heroes/creatures (you’ve made a great start here).
-   Stop destroying his own walls/rooms to dig out other rooms.
-   Use empty but claimed spaces to build new rooms. (Right now if you give the 
CP some 5x5 empty spaces to start out with he will not use those effectively.)
-   Using doors to get rooms to max efficiency.
-   Using level specials.

Original comment by Loobinex on 12 May 2015 at 12:28

[GoogleCodeExporter]

By the way, the original campaign has some maps where the CP does not dig for 
gold because he starts out with more gold than he would ever need.

Original comment by Loobinex on 12 May 2015 at 12:30

[GoogleCodeExporter]

Risk getting off-topic, but: I've got a long google doc covering what I'd like 
to do with AI, just got to get the time. :P I'm also somewhat careful to spend 
too much time discussing this as it's possible to talk forever about hobby 
projects (rather than do).

To answer the biggest of your points, nonetheless:
- Gold/digging/room build behavior is among the most extensive and difficult to 
change and make intelligent. I might be able to special code new build treasury 
room behavior to optimize for gold/gem mining. But there's so much more to take 
consideration of in reality; in particular locations of enemies and neutrals. 
DK AI back when originally written had much less CPU time to do with so it's 
not surprising it's not using any whole map analyses.
- Training is a big deal and it's way too easy to just out-train AI atm, 
agreed. All the same, I'd prefer to fix economy to the extent possible first, 
because without "fuel" there's no way to do training.
- Getting rid of crap creatures is on my TODO list, just need to be careful to 
work well with map designer constraints (don't want to drop beetles if they're 
only available creature on map, as obvious example).
- Thanks for reminding about bridge capture exploit, that should indeed not be 
so easy. As a side note, I'd have liked better if game didn't allow you to drop 
creatures on bridges as it's too easy to zero-effort pass water/lava, but I 
don't think game rule changes are very realistic (they would be debated).

I don't think there's any opposition in working towards both a better -1player 
opponent and better opponent for designed maps; for the latter one just needs 
to ensure behavior can be customized.

Original comment by eatpajdi...@gmail.com on 13 May 2015 at 9:37

[GoogleCodeExporter]

There's no opposition in working towards making both better, but having a clear 
goal helps prioritizing and focus.
I'd love to see the doc of course.

I have no objection if you want to prioritize economy over training, but 
realize two things:
1) Mefisto has recently made quite a few improvements in the economy 
department. Issue #204 is still a very important one that is still open, but 
otherwise the CP has some prety good gold management.
2) Even if you give the CP limitless gold the training is still sub-par. I 
recently did a test on DD-map Morkardar, and in Keeper95 on an unchanged map 
the creatures were better trained than in KeeperFX when I gave the CP limitless 
gold.

Original comment by Loobinex on 13 May 2015 at 10:44

[GoogleCodeExporter]

Doc is random notes in Swedish unfortunately. I'll write you a mail before I 
think of implementing/improving something so it can be discussed.

Good point regarding CP doing poorly even with limitless gold. That raises 
priority of improving training itself. I have 4 free days in a row now so 
hopefully I can do something about either the no mining bug or that, will see 
what's possible in this time span.

Original comment by eatpajdi...@gmail.com on 14 May 2015 at 10:30

[GoogleCodeExporter]

From this topic: 
https://keeperklan.com/threads/4949-Ids-Habbening/page2?p=49714&viewfull=1#post4
9714

Another map where for some unexplicable reason the CP won't mine gold. Looks 
like the same bug.

Original comment by Loobinex on 16 May 2015 at 3:20

Attachments:

• [WORK DAMN YOU.rar](https://storage.googleapis.com/google-code-attachments/keeperfx/issue-604/comment-9/WORK DAMN YOU.rar)

[GoogleCodeExporter]

On r1850 with the -newdig option it works well on both the original map and 
woudo's map.

Original comment by Loobinex on 18 May 2015 at 12:02

[GoogleCodeExporter]

you've tested these with 0.4.5 and 0.4.6, right?

Could you try a very old version, to determine whether it's a regression or bug 
which was always there?

Original comment by mefistotelis on 18 May 2015 at 9:47

[GoogleCodeExporter]

Original comment by mefistotelis on 18 May 2015 at 9:47

• Changed state: Accepted
• Added labels: Component-CompPlayer

[GoogleCodeExporter]

Bug is already there in 0.4.4.976. It is as far back as I have.

Original comment by Loobinex on 18 May 2015 at 12:59

[GoogleCodeExporter]

The bug is also there in Keeper95.exe

Original comment by Loobinex on 18 May 2015 at 11:33