def test_best_response_cfr_one_card_poker():
game = OneCardPoker.create_game(n_cards=4)
strategy, exploitabilities, strategies = cfr(game, num_iters=10, use_chance_sampling=False)
exploitability = compute_exploitability(game, strategy)
print("Exploitability: {}".format(exploitability))
assert exploitability > 0.0
def test_best_response_cfr():
"""Test we can run 10 iterations of CFR on Leduc and then compute a best
response.
"""
cards = [Card(value, suit) for value in range(3) for suit in range(2)]
game = Leduc(cards)
strategy, exploitabilities, strategies = cfr(game, num_iters=10, use_chance_sampling=False)
exploitability = compute_exploitability(game, strategy)
print("Exploitability: {}".format(exploitability))
assert exploitability > 0.0
default=1e-4,
help='The number of epochs to train the neural network for.')
parser.add_argument('--dropout_rate',
default=None,
help='The dropout rate to use.')
args = parser.parse_args()
dropout_rate = None
if args.dropout_rate:
dropout_rate = float(args.dropout_rate)
cards = get_deck(num_values=args.num_values, num_suits=args.num_suits)
game = Leduc(cards)
strategy, exploitabilities, strategies = cfr(
game,
num_iters=args.cfr_iters,
use_chance_sampling=args.use_chance_sampling)
exploitability = compute_exploitability(game, strategy)
print("Exploitability of final strategy: {}".format(exploitability))
leduc_nfsp = LeducNFSP(cards)
state_vectors = leduc_nfsp._state_vectors
state_dim = leduc_nfsp.state_dim
action_dim = leduc_nfsp.action_dim
# Now build a network.
layer_dims = [64, 64, 64]
network = build_network(state_dim,
action_dim,
layer_dims,