def test_get_symmetries():
"""Check that the function correctly creates six new reflected boards and
policy vectors, one for each axis of symmetry.
"""
game = BloomsGame(size=4, score_target=15)
board = game.getInitBoard()
# Place stones
board.place_stone(position=(3, 1), colour=1)
board.place_stone(position=(5, 1), colour=2)
board.place_stone(position=(3, 5), colour=3)
board.place_stone(position=(1, 5), colour=4)
# board.visualise(show_coords=True, title="Original")
# Create a dummy policy vector
pi = np.random.random_sample(game.getActionSize())
profiler = Profiler()
profiler.start()
symmetrical_states = game.getSymmetries(board, pi)
profiler.stop()
print(profiler.output_text(unicode=True, color=True))
assert len(symmetrical_states) == 24
assert symmetrical_states[1][1][37] == pi[52]
assert symmetrical_states[1][1][614] == pi[1277]
assert all(
[sum(x[1]) == pytest.approx(sum(pi)) for x in symmetrical_states])
def test_get_init_board():
"""Check that a new board is initialised correctly.
"""
game = BloomsGame()
board = game.getInitBoard()
assert type(board) == Board
def test_get_canonical_form_inversion():
"""Check that the board is inverted (i.e. Player -1 becomes Player 1 and
vice-a-versa) when called for Player 1.
"""
game = BloomsGame(size=4, score_target=15)
board = game.getInitBoard()
# Place stones
board.place_stone(position=(5, 1), colour=1)
board.place_stone(position=(5, 2), colour=2)
board.place_stone(position=(1, 4), colour=3)
board.place_stone(position=(1, 5), colour=4)
# Update captures
board.captures = [5, 10]
# Uncomment to visualise the inversion
# board.visualise()
board = game.getCanonicalForm(board, player=-1)
assert board.board_2d[1, 5] == 3
assert board.board_2d[2, 5] == 4
assert board.board_2d[4, 1] == 1
assert board.board_2d[5, 1] == 2
assert board.captures == [10, 5]
def test_get_action_size_base3():
"""Check that the correct action space size is returned for a base 3 board.
"""
game = BloomsGame(size=3)
n_spaces = 19
assert game.getActionSize() == 2 * n_spaces + perm(n_spaces, 2)
def test_get_game_ended_ongoing():
"""Check that the function correctly signals an ongoing game.
"""
game = BloomsGame(size=4, score_target=15)
board = game.getInitBoard()
assert game.getGameEnded(board, player=-1) == 0.0
def test_get_valid_moves():
"""Test that the valid moves vector is correctly built.
"""
game = BloomsGame(size=4)
board = game.getInitBoard()
valid_moves = game.getValidMoves(board, player=-1)
assert len(valid_moves) == game.getActionSize()
assert np.all((valid_moves == 0) | (valid_moves == 1))
def test_get_game_ended_loss():
"""Check that the function correctly signals a loss.
"""
game = BloomsGame(size=4, score_target=15)
board = game.getInitBoard()
# Set the number of captures for Player 1 (i.e. the opponent) to 15
board.captures[1] = 15
assert game.getGameEnded(board, player=-1) == -1.0
def test_get_game_ended_win():
"""Check that the function correctly signals a win.
"""
game = BloomsGame(size=4, score_target=15)
board = game.getInitBoard()
# Set the number of captures for Player -1/0 to 15
board.captures[0] = 15
assert game.getGameEnded(board, player=-1) == 1.0
def test_get_game_ended_draw():
"""Check that the function correctly signals a draw.
"""
game = BloomsGame(size=4, score_target=15)
board = game.getInitBoard()
# Fill the board
for position in board.get_empty_spaces():
board.place_stone(position=position, colour=1)
assert game.getGameEnded(board, player=-1) == pytest.approx(1e-4)
def test_init_custom():
"""Check that a new custom game is initialised correctly.
"""
game = BloomsGame(size=5, score_target=20)
assert game.size == 5
assert game.score_target == 20
def test_init_default():
"""Check that a new default game is initialised correctly.
"""
game = BloomsGame()
assert game.size == 4
assert game.score_target == 15
def test_get_next_state_valid_move():
"""Check that the board state is updated if a legal move is performed.
"""
game = BloomsGame(size=4)
board = game.getInitBoard()
# Attempt a legal move
player = -1
move = ((6, 2, 2), ())
move_idx = board.move_map_player_0[move]
next_board, next_player = game.getNextState(board, player, action=move_idx)
next_board.board_2d
for r in range(next_board.board_2d.shape[0]):
for q in range(next_board.board_2d.shape[1]):
if (q == 6) and (r == 2):
assert next_board.board_2d[r, q] == 2
else:
assert next_board.board_2d[r, q] == 0
assert next_player == -player
def test_get_next_state_invalid_move():
"""Check that the original (i.e. unchanged) board state is returned if an
invalid move is performed.
"""
game = BloomsGame(size=4)
board = game.getInitBoard()
# Initialise the board with a stone
board.place_stone((6, 2), colour=1)
initial_board_array = board.get_board_3d()
# Attempt an illegal move
player = -1
move = ((6, 2, 2), ())
move_idx = board.move_map_player_0[move]
next_board, next_player = game.getNextState(board, player, action=move_idx)
post_move_board_array = next_board.get_board_3d()
assert next_player == -player
assert np.all(initial_board_array == post_move_board_array)
def test_get_canonical_form_no_inversion():
"""Check that the original board is returned when called for Player -1.
"""
game = BloomsGame(size=4, score_target=15)
board = game.getInitBoard()
# Place stones
board.place_stone(position=(5, 1), colour=1)
board.place_stone(position=(5, 2), colour=2)
board.place_stone(position=(1, 4), colour=3)
board.place_stone(position=(1, 5), colour=4)
# Update captures
board.captures = [5, 10]
board = game.getCanonicalForm(board, player=1)
assert board.board_2d[1, 5] == 1
assert board.board_2d[2, 5] == 2
assert board.board_2d[4, 1] == 3
assert board.board_2d[5, 1] == 4
assert board.captures == [5, 10]
"""Use this script to play manually against a Blooms agent.
"""
import numpy as np
import Arena
from MCTS import MCTS
from blooms.BloomsGame import BloomsGame
from blooms.BloomsPlayers import *
from blooms.pytorch.NNet import NNetWrapper as NNet
from utils import *
# WARNING: The game size and score target should match the chosen agent
game = BloomsGame(size=5, score_target=20)
human = HumanBloomsPlayer(game).play
# WARNING: The chosen agent should match the game size and score target
model = NNet(game)
model.load_checkpoint('./notebooks/results/chkpts_board5_24hrs',
'best.pth.tar')
args = dotdict({'numMCTSSims': 100, 'cpuct': 1.0})
mcts = MCTS(game, model, args)
agent = lambda x: np.argmax(mcts.getActionProb(x, temp=0))
arena = Arena.Arena(agent, human, game)
print(arena.playGames(2, verbose=True, display=False))
def test_blooms_pytorch(self):
self.execute_game_test_random(BloomsGame(), BloomsPyTorchNNet)
def test_get_board_size():
"""Check that the correct board size is returned.
"""
game = BloomsGame()
assert game.getBoardSize() == (7, 7)
