def test_diagonal_wins():
env = MultiAgentTicTacToe()
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [0, 0], 1: [9, 0]})
env.step({0: [9, 0], 1: [1, 0]})
env.step({0: [4, 0], 1: [9, 0]})
env.step({0: [9, 0], 1: [3, 0]})
obs, rets, dones, infos = env.step({0: [8, 0], 1: [9, 0]})
assert rets[0] == 1
assert rets[1] == -1
env = MultiAgentTicTacToe()
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [2, 0], 1: [9, 0]})
env.step({0: [9, 0], 1: [1, 0]})
env.step({0: [4, 0], 1: [9, 0]})
env.step({0: [9, 0], 1: [3, 0]})
obs, rets, dones, infos = env.step({0: [6, 0], 1: [9, 0]})
assert rets[0] == 1
assert rets[1] == -1
def test_diagonal_wins_hd():
env = MultiAgentTicTacToe(20)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [0, 3], 1: [9, 3]})
env.step({0: [9, 5], 1: [1, 0]})
env.step({0: [4, 0], 1: [9, 3]})
env.step({0: [9, 6], 1: [3, 2]})
obs, rets, dones, infos = env.step({0: [8, 6], 1: [9, 3]})
assert rets[0] == 1
assert rets[1] == -1
env = MultiAgentTicTacToe(20)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [2, 6], 1: [9, 5]})
env.step({0: [9, 2], 1: [1, 8]})
env.step({0: [4, 3], 1: [9, 3]})
env.step({0: [9, 7], 1: [3, 4]})
obs, rets, dones, infos = env.step({0: [6, 2], 1: [9, 7]})
assert rets[0] == 1
assert rets[1] == -1
def test_horizontal_wins_hd():
env = MultiAgentTicTacToe(10)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [0, 9], 1: [9, 9]})
env.step({0: [9, 3], 1: [4, 9]})
env.step({0: [1, 4], 1: [9, 9]})
env.step({0: [9, 2], 1: [5, 9]})
obs, rets, dones, infos = env.step({0: [2, 4], 1: [9, 2]})
assert rets[0] == 1
assert rets[1] == -1
env = MultiAgentTicTacToe(10)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [3, 5], 1: [9, 4]})
env.step({0: [9, 7], 1: [0, 7]})
env.step({0: [4, 2], 1: [9, 8]})
env.step({0: [9, 4], 1: [1, 2]})
obs, rets, dones, infos = env.step({0: [5, 4], 1: [9, 0]})
assert rets[0] == 1
assert rets[1] == -1
env = MultiAgentTicTacToe(10)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [6, 4], 1: [9, 6]})
env.step({0: [9, 0], 1: [0, 3]})
env.step({0: [7, 6], 1: [9, 2]})
env.step({0: [9, 2], 1: [1, 3]})
obs, rets, dones, infos = env.step({0: [8, 2], 1: [9, 5]})
assert rets[0] == 1
assert rets[1] == -1
def test_vertical_wins_hd():
env = MultiAgentTicTacToe(5)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [0, 4], 1: [9, 0]})
env.step({0: [9, 2], 1: [4, 2]})
env.step({0: [3, 4], 1: [9, 3]})
env.step({0: [9, 1], 1: [5, 0]})
obs, rets, dones, infos = env.step({0: [6, 2], 1: [9, 4]})
assert rets[0] == 1
assert rets[1] == -1
env = MultiAgentTicTacToe(5)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [1, 4], 1: [9, 0]})
env.step({0: [9, 0], 1: [3, 3]})
env.step({0: [4, 1], 1: [9, 3]})
env.step({0: [9, 0], 1: [5, 0]})
obs, rets, dones, infos = env.step({0: [7, 3], 1: [9, 4]})
assert rets[0] == 1
assert rets[1] == -1
env = MultiAgentTicTacToe(5)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [2, 1], 1: [9, 2]})
env.step({0: [9, 0], 1: [3, 0]})
env.step({0: [5, 3], 1: [9, 5]})
env.step({0: [9, 0], 1: [4, 0]})
obs, rets, dones, infos = env.step({0: [8, 3], 1: [9, 3]})
assert rets[0] == 1
assert rets[1] == -1
def simulate_rollouts_tictactoe(trainers, dim):
env = MultiAgentTicTacToe(dim)
total_score = {-1: 0, 0: 0, 1: 0}
for i in range(20):
obs = env.reset()
print("--Rollout" + str(i) + "--")
print("First Player: " + str(obs[0][9]))
done = False
while not done:
#print(obs)
actions = {
i: trainer.compute_action(obs[i],
policy_id=policy_mapping_fn(i))
for i, trainer in enumerate(trainers)
}
# idk why this is suddenly necessary TODO figure that out
actions = {k: [x[0] for x in v] for k, v in actions.items()}
print(actions)
obs, rewards, dones, infos = env.step(actions)
done = dones['__all__']
if done:
total_score[rewards[0]] += 1
env.render()
print("player 0 outcomes: " + str(total_score))
def test_out_of_turn_over_not_move():
env = MultiAgentTicTacToe()
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
obs, rets, dones, infos = env.step({0: [9, 0], 1: [3, 0]})
assert rets[0] == 1
assert rets[1] == -1
def test_not_move_on_turn_hd():
env = MultiAgentTicTacToe(8)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
obs, rets, dones, infos = env.step({0: [9, 2], 1: [9, 6]})
assert rets[0] == -1
assert rets[1] == 1
def test_place_piece_on_other_piece():
env = MultiAgentTicTacToe()
obs = env.reset()
# check that it is player 0's move
assert obs[0][9] == 0
env.step({0: [4, 0], 1: [9, 0]})
obs, rets, dones, infos = env.step({0: [9, 0], 1: [4, 0]})
assert rets[0] == 1
assert rets[1] == -1
def test_out_of_turn_over_another_peice_hd():
env = MultiAgentTicTacToe(4)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [4, 3], 1: [9, 3]})
obs, rets, dones, infos = env.step({0: [5, 3], 1: [4, 3]})
assert rets[0] == -1
assert rets[1] == 1
def test_starting_state_hd():
env = MultiAgentTicTacToe(6)
obs = env.reset()
# Outer keys are players
for i in range(2):
assert len(obs[i]) == 10
# Board should appear empty to all players
for j in range(9):
assert obs[i][j] == 0
# Last position is players turn
assert obs[i][9] == 0
def test_tie_game_hd():
env = MultiAgentTicTacToe(8)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
env.step({0: [2, 4], 1: [9, 2]})
env.step({0: [9, 7], 1: [1, 3]})
env.step({0: [4, 4], 1: [9, 4]})
env.step({0: [9, 6], 1: [3, 0]})
obs, rets, dones, infos = env.step({0: [6, 7], 1: [9, 0]})
assert rets[0] == 1
assert rets[1] == -1
def test_place_high_dimension():
k = 2
env = MultiAgentTicTacToe(k)
obs = env.reset()
# check that it is plyer 0's move
assert obs[0][9] == 0
obs, rets, dones, infos = env.step({0: [4, k - 1], 1: [9, k - 1]})
# Outer keys are players
for i in range(2):
assert len(obs[i]) == 10
# Board should appear empty to all players
for j in range(9):
if j == 4:
assert obs[i][j] == 1 * k
else:
assert obs[i][j] == 0
# Last position is players turn
assert obs[i][9] == 1
obs, rets, dones, infos = env.step({0: [9, k - 1], 1: [5, k - 1]})
# Outer keys are players
for i in range(2):
assert len(obs[i]) == 10
# Board should appear empty to all players
for j in range(9):
if j == 4:
assert obs[i][j] == 1 * k
elif j == 5:
assert obs[i][j] == 2 * k
else:
assert obs[i][j] == 0
# Last position is players turn
assert obs[i][9] == 0