Description
Objectives
● Understand how fixed programs can play arbitrary games by exploring a game tree.
● Understand how to explore a game tree incrementally.
● Understand the UCT algorithm.
● Familiarize yourself with running graphical Python programs.
Requirements
● Implement uniform_bot.py that chooses amongst legal moves uniformly (use the
choice function in Python’s random library)
● Implement greedy_bot.py that chooses a legal move that maximizes the immediate
score gain (it looks 1 move into the future and chooses the one that increases its score
the most)
● Implement uct_bot.py that uses MCTS with full rollout. The reward function should be
the current player’s score minus the other player’s score.
● Implement fast_bot.py that uses MCTS to a rollout depth of at most 5.
● Use python’s time module to make sure that your bots do not think about their move
for more than a second.
● For uct_bot.py and fast_bot.py print out the number of rollouts per second at the end
of each move.
Grading Criteria
(equal weight for each question)
● Is uniform_bot.py implemented correctly?
● Is greedy_bot.py implemented correctly?
● Is uct_bot.py implemented correctly?
● Is fast_bot.py implemented correctly?
● Are rollout rates for the UCT bots visible somewhere?
Resources
● Use the Python code for UCT available on http://mcts.ai/ as a reference, but don’t just
copypaste what you find. It will take some adaptation to work with our state space and
simulator.
● Browse Cameron Browne’s MCTS lecture slides:
http://ccg.doc.gold.ac.uk/teaching/ludic_computing/ludic16.pdf
Base Code
● https://drive.google.com/a/ucsc.edu/folderview?id=0BPPiU3Ga8Z7fjlwNVJIb0pTYnp3
WEF1YlhiT0Q0MnFwZXZfX0F3LVBUZUpSa29wUWxMcnc&usp=sharing
● Commands:
○ $ python p2_gui.py
■ An interactive, graphical version of the game. Bots are disabled by
default, but you can turn them on with checkboxes and then restart the
game.
○ $ python p2_sim.py
■ A highspeed commandline simulator useful for running repeated
rounds between pairs of bots.
● To change which bots are active in either the graphical or tournament versions of the
game, edit the lines that look like this:
○ import first_bot as red_bot
● Here’s how you’d have the greedy bot play the blue player
○ import greedy_bot as blue bot
● In your bot modules, implement a function called “think” that takes two arguments:
○ state:
■ The current state of the game (an instance of State from p2_game.py).
Your code should not depend on the internals of the state object at all.
Here is the allowed interface:
● state.get_whos_turn() → ‘red’ or ‘blue’
● state.get_score() → dict of scores for each player
● state.get_moves() → a list of legal moves
● state.apply_move(move) → updates state to apply a move,
assuming it was legal
● state.copy() → produces a copy of this state that can be mutated
(with apply_move) without changing the original
○ quip(string):
■ A function you can call with a string to show text in the GUI or print it in
the command line for the highspeed simulator