cangiuli / hanabi Goto Github PK

The datatype play is currently

  datatype play = Discarded of card
                | Played of card
                | HintedSuit of player * suit * card list
                | HintedRank of player * rank * card list

This has two problems:

• It is not clear which card is played/discarded when a player has two copies of the same card in their hand.
• The information which was known about the played/discarded card is lost.

I propose to replace the occurrences of card by the rank of that card in the hand. Also, in Played and Discarded add fields of type info list which represents all the clues given to that player while that card was in their hand. Concretely, the datatype will be

  datatype play = Discarded of int * info list
                | Played of int * info list
                | HintedSuit of player * suit * int list
                | HintedRank of player * rank * int list

This has as extra advantage that the fields HintedSuit/HintedRank don't have to be modified for the player Me before passing the information.

A different thing: I think it would also simplify things a little bit if datatype player (which is currently Me | Other of int is just int, where 0 is Me and i+1 is Other i)

If you agree, I can make these changes (or if you want you can do them, that's of course also fine, but then please first merge my branch personal to avoid merge conflicts).

Many functions currently in SimplePlayer will be shared among all PLAYERs. These should be factored out into another structure.

To compare performance of a new feature, it would be nice to run multiple games and see if the average score increases. If you run multiple games, you probably don't want the game log, but only the achieved score (or maybe only the average score of all games). It would be nice if we can just add a flag to the terminal command to specify how much games we want to play (by default 1).

I tried adding a ref representing the memory of the players here: 52297b3
This is a summary of my approach:

  val play : unit -> state -> action =
  let
    val m : memory ref = ref initialMemory
  in
    fn u => fn s =>
    (...)
  end

I have two problems:

• I want initialMemory to depend on the number of players, but it cannot take (s : state) as argument yet.
• The memory should revert to initialMemory between games, but I don't know how I can achieve that.

Any ideas?

Although I would like to SimplePlayer to remain fairly straightforward, there is clearly much room for improvement:

• Keep track of extra information implied by save clues, and act on it
• Can discard clued cards when they're guaranteed to be useless (discardOldestUnclued)
• A card is also useless when unreachable (isUseless)
• Several 5's usually end up in the discard right now

We can also collect ideas for a more complicated player:

• Finesses
• Reverse finesses
• Bluffs
• Awareness of the endgame