kaiks / unobot Goto Github PK

Implement a command that would

1. download files from a given location (in a zip format)
2. unpack them into the unobot folder
3. load all files in the folder into the currently running program

Bonus:
4. implement version (like in ZbojeiJureq)
5. implement transactional loading:
a) make copies of all files
b) try to load a file
c) if SyntaxError is raised:

1. unload all files that have not been previously loaded
2. load previous versions of new files
3. abort further loading
   d) loader should never be overwritten

find a way with cinch to log caught exceptions

• Adjust the sample size (10000 randomized hands?)
• check for equivalent plays, e.g. all else being equal wr -> r4 = wr -> r6
  though this is probably more complicated (later plays can affect the tree), so check that later when in trouble
• Opponent should probably play with the naive strategy, esp. if it's worse performance wise

The algorithm would look like that:

1. sample N hands from the stack pile (discarding by pe'd colors!, discarding by figures)
2. check immediately playable cards
3. for each playable card, create a minmax node
   3.5) final node score is winning probability
4. choose the best node

Adversary card count is set to 6 when unobot starts the game because of this message:

[22:58:56] unobot's turn. Top card: [7]

Should be 7 cards.

When unobot goes second, the card count is correct.

bot responds to 1 message late with jruby 9k

The bot chokes at 8 cards, which should not be the case.

Even if we do 8! operations, that makes 40k and should not be taking more than 15 seconds.

E.g.: password protected host-based session that expires after a given amount of time

We need more and better tests to test for breaking changes and drive code refactoring

Need to implement a command that would save a game's history (possibly with debug messages) in a file and notify me

Bot gets lost when he has 8+ cards. E.g., it's not aggressive enough

Currently the bot only looks at probabilities based on cards that have not been played. It should also take into account what cards caused the adversary to pick a card.

e.g.:
bot has: y6, y8, ys, rs, r8, w
top card: r6
pl y6
pe
pl g6
bot should play wy rather than wr, because he knows that a doesn't have yellow cards.

once we know that the enemy doesn't have color X, we should remove it from potential card stack (for this turn at least)

same with figures

1. use sorbet or contracts for types
2. add tests
3. check for other opportunities for refactoring

sometimes, the card count is not accurate

To include:

• channels
• networks
• host nicks
  etc

in order to be able to more accurately simulate possible worlds when heuristically evaluating plays

Playing a skip reduces adversary's card count. It should not.

W gets played before WD4 when adversary has only one card. Should most likely always be the other way around.