2048-agent

Create and run agents to play Gabriele Cirulli's 2048!

Get running

You'll need Firefox and a Ruby interpreter (ruby, jruby, etc)

git clone https://github.com/rmseng/2048-agent.git
bundle install
ruby twenty_forty_eight.rb --help
ruby twenty_forty_eight.rb -a GreedyAgent -t 2

Get developing

Create a file for your new agent in the agents directory
Implement the agent with:

initialize with any setup your agent needs
next_move that takes the gameboard as its parameter and returns your agent's move

Check greedy_agent.rb or sample_agent.rb in the agents directory for examples. 3. ruby twenty_forty_eight.rb -a <YourAgent> 4. Have fun!

Tips

TwentyFortyEight can help you run parallel trials for your agent, monitor its decisions, and even run it locally for a big speed boost. See your options: ruby twenty_forty_eight.rb --help

TwentyFortyEight::Gameboard comes with many useful methods. Some of these methods return or yield TwentyFortyEight::Gameboard::Tile objects, which have attributes row, column, and value (which use zero-based coordinates.) Here are some of Gameboard's useful methods:

size self-explanatory
score self-explanatory
valid_moves returns [:left, :right, :up, :down]
move(direction, insert_random = false) returns a new copy of the gameboard with the move simulated (with or without adding a new random tile)
would_tiles_move? tells you if a move will change the game state
== compares tile values and positions (not scores)
board_full? returns true even if there is a possible combination to be made
won? self-explanatory
lost? self-explanatory
tiles enumerates each Tile
horizontal_pairs enumerates horizontal pairs of Tiles
vertical_pairs enumerates vertical pairs of Tiles
adjacent_tiles(row, col) enumerates the 2, 3, or 4 immediately adjacent Tiles
largest_tile_value returns the value of the highest-valued tile
to_s prints a ASCII representation of the gameboard
free_cells returns an array of Tiles with nil (empty) values
get_value(row, col) returns either nil or an integer, depending on the presence and value of a Tile

Starting a new trial means a new agent will be instantiated. If you want to write a learning agent, you could use class variables, but be sure to synchronize them if you're testing them in parallel.

After you think you've got an agent worthy of extensive testing, try using the 'local' option to simulate many games locally (without running Firefox.) TwentyFortyEight will simulate games of 2048 and collect the results for you.

Contributing

If you find a bug, please submit a pull request - I'll happily confirm and merge it.

Notes