Update.

[picoclvr.git] / main.py

diff --git a/main.py b/main.py
index 3db87df..6e8ebff 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -92,6 +92,17 @@ parser.add_argument("--maze_width", type=int, default=21)
 
 parser.add_argument("--maze_nb_walls", type=int, default=15)
 
+##############################
+# Snake options
+
+parser.add_argument("--snake_height", type=int, default=6)
+
+parser.add_argument("--snake_width", type=int, default=8)
+
+parser.add_argument("--snake_nb_colors", type=int, default=3)
+
+parser.add_argument("--snake_length", type=int, default=100)
+
 ######################################################################
 
 args = parser.parse_args()
@@ -623,6 +634,75 @@ class TaskMaze(Task):
 
 ######################################################################
 
+
+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)
+    count = 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
+        i = torch.arange(val.size(0), device=device)
+        j = val.argmax(1)
+        snake_direction = snake_next_direction[i, j]
+
+        sequences[:, 2 * l] = worlds[count, snake_position[:, 0], snake_position[:, 1]]
+        sequences[:, 2 * l + 1] = snake_direction
+
+        # nb x 2
+        snake_position = snake_next_position[i, j]
+
+    return sequences, worlds
+
+    # print(snake_position)
+
+
+# generate_snake_sequences(nb=1, height=4, width=6, nb_colors=3, length=20)
+# exit(0)
+
+
 class TaskSnake(Task):
     def __init__(
         self,
@@ -631,7 +711,8 @@ class TaskSnake(Task):
         batch_size,
         height,
         width,
-        nb_walls,
+        nb_colors,
+        length,
         device=torch.device("cpu"),
     ):
         self.batch_size = batch_size
@@ -639,10 +720,14 @@ class TaskSnake(Task):
         self.width = width
         self.device = device
 
-        # self.train_input = 
-        # self.test_input = 
+        self.train_input, self.train_worlds = generate_snake_sequences(
+            nb_train_samples, height, width, nb_colors, length, self.device
+        )
+        self.test_input, self.test_worlds = generate_snake_sequences(
+            nb_test_samples, height, width, nb_colors, length, self.device
+        )
 
-        self.nb_codes = max(self.train_input.max(), self.train_input.max()) + 1
+        self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
 
     def batches(self, split="train", nb_to_use=-1, desc=None):
         assert split in {"train", "test"}
@@ -656,6 +741,9 @@ class TaskSnake(Task):
         ):
             yield batch
 
+    def vocabulary_size(self):
+        return self.nb_codes
+
 
 ######################################################################
 
@@ -708,6 +796,18 @@ elif args.task == "maze":
         device=device,
     )
 
+elif args.task == "snake":
+    task = TaskSnake(
+        nb_train_samples=args.nb_train_samples,
+        nb_test_samples=args.nb_test_samples,
+        batch_size=args.batch_size,
+        height=args.snake_height,
+        width=args.snake_width,
+        nb_colors=args.snake_nb_colors,
+        length=args.snake_length,
+        device=device,
+    )
+
 else:
     raise ValueError(f"Unknown task {args.task}")