Re: [Pio-temporary] Ideas for AI playing development cards

[Roland C. <rc...@us...>]

On Sunday 02 October 2005 22:21, Brian Wellington wrote:
> On Sat, 1 Oct 2005, Roland Clobus wrote:
It's now a few weeks later, but I haven't stopped thinking about this, 
I was working on the MinGW port.

> > Year of Plenty:
> > Current: will be played anytime the bank has at least 2 resources. 
If
> > the bank does not contain anything the AI likes, it will request 
the
> > resources for a road, and if the bank is too empty, it will just 
pick
> > something at random.
> >
> > Proposed: the bank must have at least 2 resources. If the bank 
does
> > not contain anyting the AI likes, determine whether it is possible 
to
> > get enough resources to perform a maritime trade to get the 
resource
> > the AI needs (in the next turn). If not, determine whether the AI 
is
> > allowed to place a road [*1] and the resources are in the bank, 
and
> > request that. If all fails: don't play this card yet.
> 
> I'm a bit confused by this logic.  If the bank doesn't contain any 
> resources the AI wants, it doesn't matter whether the AI can get 
enugh 
> resources to be able to perform a maritime trade next turn, since 
there's 
> a good chance there still won't be any resources it wants, and it 
also 
> means the AI is more likely to have over 7 cards.

You're right. The AI takes two risks with that scenario. 

> I wonder if it would be possible to figure out if the AI is 1-2 
cards 
> short of a city/settlement/road it wants to build, and only play the 
card 
> in that case.

New proposal: the bank must have at least 2 resources. Determine 
whether the AI needs one or two cards to buy something, and the 
resources are in the bank. If not, determine whether the AI is 
allowed to place a road [*1] and the resources are in the bank, and
request that. If all fails: don't play this card yet.

> > Monopoly:
> > Current: will be played when the other players have at least six
> > resources cards in total. If the AI doesn't need anything in
> > particular, it will pick the resource that all players have the 
most
> > of.
> >
> > Proposed:  will be played when the other players have at least six
> > resources cards in total. If the AI doesn't need anything in
> > particular, don't play it yet.
> 
> Agreed, although I wonder if the AI should take into account the 
number of 
> cards.  For example, if the AI wants a lumber, and will only get 1 
lumber 
> by playing monopoly, but could get 10 brick, taking the brick would 
be a 
> much better idea, since it could do a 4:1 (or better) trade and 
still come 
> out way ahead.

I think the minimum number of resources cards in the hands of the 
other players was added to prevent the monopoly card to be played too 
early. If the AI would gain only one card, would it be worth it to 
play the card? (I think only when it will result in the final victory 
point)

Because the AI can see how many resources are still in the bank, 
it can make an accurate guess of the profits of playing the card.
If the AI wants a specific resource, it could alternatively try to get 
it from maritime trade, after playing the Monopoly card for the 
abundant resource (but it must be sure that it gets enough cards to 
perform the trade and to buy something).

> > Gold:
> > Current: when the AI does want something, get it. Otherwise, start
> > emptying the bank starting from the first resource.
> >
> > Proposed: when the AI does want something, get it. Otherwise, 
start
> > emptying the bank starting from the resource that has the best
> > maritime trade ratio.
> 
> Unlike the dev cards, gold can't be deferred until later, so there 
might 
> be situations where the AI doesn't want anything because it can, for 
> example, already build a settlement.  It might be worth it to figure 
out 
> what the AI will do first, and if its first action is to buy 
something, 
> re-evaluate what it would want after buying it, then choose that, 
falling 
> back to something like your suggestion afterwards.  This might be 
too 
> complicated, though, and your idea would still help.

Sounds like a good idea. It is more complicated though.
Perhaps this can be changed in two steps.

Roadbuilding: see task #120371.

Regards,
Roland Clobus