def game5():
situation_size, message_size, prediction_size, func_size, hidden_size = (
10,
2,
10,
20,
64,
)
game = Game(situation_size, message_size, prediction_size, func_size,
hidden_size, 1.2)
game.play()
plot_messages_information(game, 40)
def game3b():
situation_size, message_size, prediction_size, func_size, hidden_size = (
10,
2,
2,
4,
64,
)
game = Game(situation_size, message_size, prediction_size, func_size,
hidden_size)
game.play()
plot_messages_information(game)
predict_information_from_messages(game)
clusterize_messages(game)
def game3():
situation_size, information_size, message_size, prediction_size, hidden_sizes = (
10,
4,
2,
2,
(64, 64),
)
game = Game(
situation_size,
information_size,
message_size,
prediction_size,
hidden_sizes,
use_context=True,
)
game.play()
plot_messages_information(game, 40)
predict_information_from_messages(game)
clusterize_messages(game)
def game7():
situation_size, message_size, prediction_size, func_size, hidden_size = (
10,
2,
10,
4,
64,
)
game = Game(situation_size, message_size, prediction_size, func_size,
hidden_size)
print_first = True
for lr in [0.01, 0.001, 0.0001]:
play_game(game, 1000, learning_rate=lr)
if print_first:
logging.info(
f"Epoch {game.loss_per_epoch[0][0]}:\t{game.loss_per_epoch[0][1]:.2e}"
)
print_first = False
logging.info(
f"Epoch {game.loss_per_epoch[-1][0]}:\t{game.loss_per_epoch[-1][1]:.2e}"
)
plot_messages_information(game, 40)
# Compute the average messages
game.average_messages(100)
replications_per_func = 10
situations = torch.randn(replications_per_func * game.func_size,
game.situation_size)
func_switches = torch.cat(
[torch.arange(game.func_size) for _ in range(replications_per_func)])
targets = game._target(situations, func_switches)
LOSS = game.criterion(game.discrete_forward(situations, func_switches),
targets).item()
logging.info(f"Loss: {LOSS:.2e}")