Update.
authorFrançois Fleuret <francois@fleuret.org>
Tue, 6 Feb 2024 14:15:21 +0000 (15:15 +0100)
committerFrançois Fleuret <francois@fleuret.org>
Tue, 6 Feb 2024 14:15:21 +0000 (15:15 +0100)
grid.py
main.py
tasks.py

diff --git a/grid.py b/grid.py
index 2135710..1287ad5 100755 (executable)
--- a/grid.py
+++ b/grid.py
@@ -21,14 +21,18 @@ class GridFactory:
         nb_questions=4,
         nb_shapes=6,
         nb_colors=6,
+        nb_play_steps=3,
     ):
         assert size % 2 == 0
         self.size = size
         self.max_nb_items = max_nb_items
         self.max_nb_transformations = max_nb_transformations
         self.nb_questions = nb_questions
+        self.nb_play_steps = nb_play_steps
         self.name_shapes = ["A", "B", "C", "D", "E", "F"]
         self.name_colors = ["red", "yellow", "blue", "green", "white", "purple"]
+        self.vname_shapes = ["vA", "vB", "vC", "vD", "vE", "vF"]
+        self.vname_colors = ["vred", "vyellow", "vblue", "vgreen", "vwhite", "vpurple"]
 
     def generate_scene(self):
         nb_items = torch.randint(self.max_nb_items - 1, (1,)).item() + 2
@@ -42,33 +46,77 @@ class GridFactory:
         shp = shp[i]
         return col.reshape(self.size, self.size), shp.reshape(self.size, self.size)
 
-    def random_transformations(self, scene):
+    def random_object_move(self, scene):
         col, shp = scene
+        while True:
+            a = (col.flatten() >= 0).nonzero()
+            a = a[torch.randint(a.size(0), (1,)).item()]
+            i, j = a // self.size, a % self.size
+            assert col[i, j] >= 0
+            dst = [(i, j), (i - 1, j), (i + 1, j), (i, j - 1), (i, j + 1)]
+            dst = list(
+                filter(
+                    lambda x: x[0] >= 0
+                    and x[1] >= 0
+                    and x[0] < self.size
+                    and x[1] < self.size
+                    and col[x[0], x[1]] < 0,
+                    dst,
+                )
+            )
+            if len(dst) > 0:
+                ni, nj = dst[torch.randint(len(dst), (1,)).item()]
+                col[ni, nj] = col[i, j]
+                shp[ni, nj] = shp[i, j]
+                col[i, j] = -1
+                shp[i, j] = -1
+                break
+
+        return col, shp
 
+    def transformation(self, t, scene):
+        col, shp = scene
+        if t == 0:
+            col, shp = col.flip(0), shp.flip(0)
+            description = "<chg> vertical flip"
+        elif t == 1:
+            col, shp = col.flip(1), shp.flip(1)
+            description = "<chg> horizontal flip"
+        elif t == 2:
+            col, shp = col.flip(0).t(), shp.flip(0).t()
+            description = "<chg> rotate 90 degrees"
+        elif t == 3:
+            col, shp = col.flip(0).flip(1), shp.flip(0).flip(1)
+            description = "<chg> rotate 180 degrees"
+        elif t == 4:
+            col, shp = col.flip(1).t(), shp.flip(1).t()
+            description = "<chg> rotate 270 degrees"
+
+        return (col.contiguous(), shp.contiguous()), description
+
+    def random_transformations(self, scene):
         descriptions = []
         nb_transformations = torch.randint(self.max_nb_transformations + 1, (1,)).item()
         transformations = torch.randint(5, (nb_transformations,))
 
         for t in transformations:
-            if t == 0:
-                col, shp = col.flip(0), shp.flip(0)
-                descriptions += ["<chg> vertical flip"]
-            elif t == 1:
-                col, shp = col.flip(1), shp.flip(1)
-                descriptions += ["<chg> horizontal flip"]
-            elif t == 2:
-                col, shp = col.flip(0).t(), shp.flip(0).t()
-                descriptions += ["<chg> rotate 90 degrees"]
-            elif t == 3:
-                col, shp = col.flip(0).flip(1), shp.flip(0).flip(1)
-                descriptions += ["<chg> rotate 180 degrees"]
-            elif t == 4:
-                col, shp = col.flip(1).t(), shp.flip(1).t()
-                descriptions += ["<chg> rotate 270 degrees"]
-
-            col, shp = col.contiguous(), shp.contiguous()
-
-        return (col, shp), descriptions
+            scene, description = self.transformation(t, scene)
+            descriptions += [description]
+
+        return scene, descriptions
+
+    def visual_scene2str(self, scene):
+        col, shp = scene
+        r = []
+        for i in range(self.size):
+            s = []
+            for j in range(self.size):
+                if col[i, j] >= 0:
+                    s += [self.vname_colors[col[i, j]], self.vname_shapes[shp[i, j]]]
+                else:
+                    s += ["v_", "v+"]
+            r += s  # .append(" ".join(s))
+        return " ".join(r)
 
     def print_scene(self, scene):
         col, shp = scene
@@ -148,8 +196,22 @@ class GridFactory:
 
         return properties
 
+    def generate_scene_and_play(self):
+        scene = self.generate_scene()
+        steps = [self.visual_scene2str(scene)]
+        for t in range(self.nb_play_steps - 1):
+            if torch.randint(4, (1,)).item() == 0:
+                scene, _ = self.transformation(torch.randint(5, (1,)), scene)
+            else:
+                scene = self.random_object_move(scene)
+            steps.append(self.visual_scene2str(scene))
+        return " | ".join(steps)
+
     def generate_scene_and_questions(self):
         while True:
+            # We generate scenes until we get one with enough
+            # properties
+
             while True:
                 start_scene = self.generate_scene()
                 scene, transformations = self.random_transformations(start_scene)
@@ -157,6 +219,12 @@ class GridFactory:
                 if len(true) >= self.nb_questions:
                     break
 
+            # We generate a bunch of false properties by shuffling the
+            # scene and sometimes adding properties from totally
+            # different scenes. We try ten times to get enough false
+            # properties and go back to generating the scene if we do
+            # not succeed
+
             for a in range(10):
                 col, shp = scene
                 col, shp = col.view(-1), shp.view(-1)
@@ -172,6 +240,7 @@ class GridFactory:
                 # We sometime add properties from a totally different
                 # scene to have negative "there is a xxx xxx"
                 # properties
+
                 if torch.rand(1).item() < 0.2:
                     other_scene = self.generate_scene()
                     false += self.all_properties(other_scene)
@@ -199,15 +268,18 @@ class GridFactory:
 
         return start_scene, scene, result
 
-    def generate_samples(self, nb, progress_bar=None):
+    def generate_samples(self, nb, fraction_play=0.0, progress_bar=None):
         result = []
 
-        r = range(nb)
+        play = torch.rand(nb) < fraction_play
         if progress_bar is not None:
-            r = progress_bar(r)
+            play = progress_bar(play)
 
-        for _ in r:
-            result.append(self.generate_scene_and_questions()[2])
+        for p in play:
+            if p:
+                result.append(self.generate_scene_and_play())
+            else:
+                result.append(self.generate_scene_and_questions()[2])
 
         return result
 
@@ -238,4 +310,14 @@ if __name__ == "__main__":
     print()
     print(questions)
 
+    # print(grid_factory.visual_scene2str(scene))
+
+    # grid_factory.print_scene(scene)
+    # for t in range(5):
+    # scene = grid_factory.random_object_move(scene)
+    # print()
+    # grid_factory.print_scene(scene)
+
+    print(grid_factory.generate_scene_and_play())
+
 ######################################################################
diff --git a/main.py b/main.py
index 69731ff..9f82594 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -104,6 +104,8 @@ parser.add_argument("--rpl_no_prog", action="store_true", default=False)
 
 parser.add_argument("--grid_size", type=int, default=6)
 
+parser.add_argument("--grid_fraction_play", type=float, default=0)
+
 ##############################
 # picoclvr options
 
@@ -554,6 +556,7 @@ elif args.task == "grid":
         nb_test_samples=args.nb_test_samples,
         batch_size=args.batch_size,
         size=args.grid_size,
+        fraction_play=args.grid_fraction_play,
         logger=log_string,
         device=device,
     )
index a53d213..08aa8ca 100755 (executable)
--- a/tasks.py
+++ b/tasks.py
@@ -1475,6 +1475,7 @@ class Grid(Task):
         nb_test_samples,
         batch_size,
         size,
+        fraction_play=0.0,
         logger=None,
         device=torch.device("cpu"),
     ):
@@ -1490,10 +1491,12 @@ class Grid(Task):
             )
 
         self.train_descr = self.grid_factory.generate_samples(
-            nb_train_samples, lambda r: tqdm.tqdm(r)
+            nb=nb_train_samples,
+            fraction_play=fraction_play,
+            progress_bar=lambda r: tqdm.tqdm(r),
         )
         self.test_descr = self.grid_factory.generate_samples(
-            nb_test_samples, lambda r: tqdm.tqdm(r)
+            nb=nb_test_samples, fraction_play=0.0, progress_bar=lambda r: tqdm.tqdm(r)
         )
 
         # Build the tokenizer