coreylowman / hanabi-mcts Goto Github PK

That would help differentiate between a card just drawn with no hints, and a card you've had for 5 turns with no hints.

Given action and expected opponent action (e.g. hint Suit 1 and play the hinted card), if the opponent doesn't do it, reduce probability of that action pair.

Based on how many cards are left in deck. E.g. if all the Red 3s are gone, then you will miss out on 2 points in the final score.

Would help #3

Related to #9

A hint of yellow tells you that all the cards that don't match are not yellow

A determinization that is lower probability should be weighted less than a determinization that is higher probability...

This will also encourage using hints more to reduce uncertainty.

But is this already handled by the fact that a determinization that is low probability will already be sampled less?

Only extracting the other players hand might not produce super consistent samples...

Both probably need to be extracted at the start of a rollout. Which basically means don't re deferminize every turn?

• don't play a no hints card
• don't play a card that will use a black token
• don't discard a hinted card that is usable?
• don't hint that would add 0 info