def test_seq2slate_transformer_simulation_hard_tsp(self):
"""
Solve Traveling Salesman Problem. Data comes from multiple sets of cities.
"""
device = torch.device("cuda")
batch_size = 4096
epochs = 8
num_batches = 300
expect_reward_threshold = 1.02
hidden_size = 32
num_candidates = 6
diverse_input = True
learning_rate = 0.001
learning_method = SIMULATION
policy_gradient_interval = 1
run_seq2slate_tsp(
MODEL_TRANSFORMER,
batch_size,
epochs,
num_candidates,
num_batches,
hidden_size,
diverse_input,
learning_rate,
expect_reward_threshold,
learning_method,
policy_gradient_interval,
device,
)
def test_seq2slate_transformer_off_policy_simple_tsp(self):
"""
Solve Traveling Salesman Problem. Data comes from one set of nodes (cities).
"""
device = torch.device("cpu")
batch_size = 4096
epochs = 1
num_batches = 100
expect_reward_threshold = 1.02
hidden_size = 32
num_candidates = 6
diverse_input = False
learning_rate = 0.001
learning_method = OFF_POLICY
policy_gradient_interval = 1
run_seq2slate_tsp(
MODEL_TRANSFORMER,
batch_size,
epochs,
num_candidates,
num_batches,
hidden_size,
diverse_input,
learning_rate,
expect_reward_threshold,
learning_method,
policy_gradient_interval,
device,
)
