+
+def generate_snake_sequences(
+ nb, height, width, nb_colors, length, device=torch.device("cpu")
+):
+ worlds = torch.randint(nb_colors, (nb, height, width), device=device)
+ # nb x 2
+ snake_position = torch.cat(
+ (
+ torch.randint(height, (nb, 1), device=device),
+ torch.randint(width, (nb, 1), device=device),
+ ),
+ 1,
+ )
+ snake_direction = torch.randint(4, (nb,), device=device)
+ sequences = torch.empty(nb, 2 * length, device=device, dtype=torch.int64)
+ i = torch.arange(nb, device=device) # [:,None]
+
+ for l in range(length):
+ # nb x 3
+ snake_next_direction = torch.cat(
+ (
+ (snake_direction[:, None] - 1) % 4,
+ snake_direction[:, None],
+ (snake_direction[:, None] + 1) % 4,
+ ),
+ 1,
+ )
+
+ # nb x 3
+ vh = (snake_next_direction + 1) % 2 * (snake_next_direction - 1)
+ vw = snake_next_direction % 2 * (snake_next_direction - 2)
+
+ # nb x 3 x 2
+ snake_next_speed = torch.cat((vh[:, :, None], vw[:, :, None]), 2)
+ snake_next_position = snake_position[:, None, :] + snake_next_speed
+
+ # nb x 3
+ val = torch.logical_and(
+ torch.logical_and(
+ snake_next_position[:, :, 0] >= 0, snake_next_position[:, :, 0] < height
+ ),
+ torch.logical_and(
+ snake_next_position[:, :, 1] >= 0, snake_next_position[:, :, 1] < width
+ ),
+ ).float()
+ val = (
+ torch.rand_like(val) * val * torch.tensor([[1.0, 4.0, 1.0]], device=device)
+ )
+
+ # nb
+ j = val.argmax(1)
+ snake_direction = snake_next_direction[i, j]
+
+ sequences[:, 2 * l] = worlds[i, snake_position[:, 0], snake_position[:, 1]] + 4
+ sequences[:, 2 * l + 1] = snake_direction
+
+ # nb x 2
+ snake_position = snake_next_position[i, j]
+
+ return sequences, worlds
+
+
+# generate_snake_sequences(nb=1, height=4, width=6, nb_colors=3, length=20)
+# exit(0)
+
+