This project implements a card game where players are controlled by AI agents using reinforcement learning. The agents learn to play cards to maximize their points while managing their resources.
To run the game, you will need Python 3.x installed on your system.
- Clone the repository to your local machine.
- Install the required dependencies by running
pip install -r requirements.txt.
To start the game, run the main.py script from the command line:
python main.py
The AI agents can be trained by running the train_model function within ui.py. This function takes the number of episodes to train as an argument.
from ui import train_model
train_model(root, num_agents_entry, num_episodes_entry, randomize_agents_var)
Below is a list of game functions that are handled:
- Recalling workers for each season
- Meadow
- Hands
- Drawing cards in summer
- Real card names, points, cost
- 15 card city limit, including for the fool
- Unique cards
- Basic locations
- Prosperity cards
Below is a list of game functions that are not yet handled: V1 Plan:
- Real card rules that affect other cards in play
- Real card rules that add a worker location
- Real card rules that activate when a card is played
- Both harvests
- Basic Events
- Shared locations
- Forest locations
- Special Events
- King card rules
- Haven
- Journey
- Occupation
- Occupation lock
- Open Destination cards
- Add ID# and details of each card to game state (if needed?)
- When testing, pause after each draw to select the correct cards into all positions and wait for a continue button V2 Plan:
- play_card currently prioritizes taking cards from the meadow, but we need the ai to choose either the one from the hand or the meadow
- Gatherers should only add points for pairs with Harvesters, not just any lone Harvester.
- Similarly, harvesters should only grant resources when paired.
- Extra locations when there are 4 players
- Discarding cards that would be over the hand limit when donating cards to an opponent
- Drawing all cards from the meadow before replenishing when multiple cards are drawn from the meadow
- If a player has passed, they can not be given any resources or cards
- In the case of ties, the player with more events wins, if there is still a tie, then count resources
- Re-shuffling an empty deck in training mode
- The Chapel and Shepherd are not implemented yet because they are annoying
- AI could decide which person to give the fool to (currently just picks the next player)
- The AI does not handle checking for if the opponent already has a fool and might give them 2 fools which is a unique card
- The AI currently checks that all opponents have free space for a fool, it only needs one
- The crane currently only reduces the cost of resources starting with stone and any other resources if there is any leftover, the AI should be able to choose which of any combination of the resources to reduce
- The AI currently has a fixed setting to choose the next resource on the list to replace another resource with when using the judge. Ideally the AI would should be able to choose which resource to swap if there is more than one option.
- The AI currently prioritizes using innkeepers, then cranes, then judges, but ideally it should be able to choose between these. This also skews the stats towards innkeepers and away from judges.
- The AI currently only uses the judge when it has to, but in theory it could choose to use the judge even when it could otherwise afford the card
Contributions to this project are welcome. Please fork the repository and submit a pull request with your changes.