def test_random_noughts_and_crosses_player_gives_equal_action_probabilities():
nac = NoughtsAndCrosses()
player = RandomPlayer(game=nac)
action, action_probs = player.choose_action(nac.initial_state,
return_probabilities=True)
next_states = nac.legal_actions(nac.initial_state)
expected_action_probs = {
action: 1 / len(next_states)
for action in next_states.keys()
}
for action in expected_action_probs.keys():
np.testing.assert_almost_equal(action_probs[action],
expected_action_probs[action])
def test_mcts_noughts_and_crosses_player_gives_optimal_moves(
state, optimal_actions):
# seed the random number generator.
np.random.seed(0)
nac = NoughtsAndCrosses()
estimator = create_trivial_estimator(nac)
player = MCTSPlayer(game=nac,
estimator=estimator,
mcts_iters=100,
c_puct=0.5,
tau=1)
action, action_probs = player.choose_action(state,
return_probabilities=True)
print(action_probs)
assert max(action_probs, key=action_probs.get) in optimal_actions
"""This program plays noughts and crosses using Monte Carlo Tree Search and a
trivial evaluator. For nonterminal states, the evaluator returns the uniform
probability distribution over available actions and a value of 0. In a terminal
state, we back up the utility returned by the game.
"""
import numpy as np
from alphago.games.noughts_and_crosses import NoughtsAndCrosses
from alphago.estimator import create_trivial_estimator
from alphago.player import MCTSPlayer
if __name__ == "__main__":
nac = NoughtsAndCrosses()
evaluator = create_trivial_estimator(nac.legal_actions)
state = nac.INITIAL_STATE
computer_player_no = np.random.choice([1, 2])
computer_player = MCTSPlayer(nac,
evaluator,
mcts_iters=2000,
c_puct=0.5,
tau=0.01)
human_player_no = 1 if computer_player_no == 2 else 2
print("You are player: {}".format(human_player_no))
while not nac.is_terminal(state):
player_no = nac.current_player(state)
next_states = nac.legal_actions(state)
if player_no == computer_player_no:
action = computer_player.choose_action(state)
computer_player.update(action)
print("Taking action: {}".format(action))
