Collectible Card Game Headquarters

Not necessarily bugs but a sticky topic to highlight what appears to be stupid AI (or human!) plays might be useful. Here is one play I do not understand - Cunning Sparkmage's ability is activated and instead of taking out the 1/1 Snake token it hits the 2/2 Wolf token instead for an impotent 1?

AI is MCTS (cheat), level 4.

Thought I saw what was going on, but no... No idea why

I have added a new test scenario that will let you replicate this - see https://groups.google.com/forum/#!topic ... jnVY0Vfb88.

not strictly related, but I thought this would be the est topic for it: AI is grafting nearly every creature I play. while I'm grateful for it, unless you have a specific deck, it's really an unwise action to do.

jerichopumpkin wrote:not strictly related, but I thought this would be the est topic for it: AI is grafting nearly every creature I play. while I'm grateful for it, unless you have a specific deck, it's really an unwise action to do.

I think this is probably due to the choice not having an implicit target picker (there is no real target). The AI is essentially choosing blind. "Adding a +1/+1 counter is surely a good thing"

Which AI are you using? I mainly use Monty Carlo, and didn't encounter this when implementing graft.

it was Minimax

jerichopumpkin wrote:it was Minimax

I may be totally wrong (Melvin's the AI master), but I think that MiniMax doesn't look so far into the future.

I'll see if there's some way of having the AI only agreeing to graft it's own creatures.

I have given up on MCTS and have been playing the las vegas Ai for a while now and I think this plays a pretty good aggressive game but still sometimes it just is so...

What makes this even more incomprehensible/ironic is that Serra Ascendant is the card being played and which positively wants you to increase your life (to 30).

Is this why you get suicide attacks.

Situation ai has 1/1 player has 2/2. it is normal for the ai to attack into the 2/2 and just lose the creature.

reasoning. ai hmm if I attack there are two outcomes they block I lose no life and they don't block they lose 1 life. If I don't attack no one loses life. Therefore attack is better.

If this is the case it should be possible to test it and fix it.
Of course adding an evaluation function that doesn't solely depend on life totals wld be better using permanents in play on both sides and cards in hand for example in addition to life.

Also does the ai still assume that cards in the opps hand are 8 mana 1/1 first striking hasty fliers cos if so there wld seem to be room for an improvement here for example assuming all cards are 2/2 creatures for 2 mana has to be better even.