X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=reasoning.py;h=c545e97f0b6517d7bbec6d163a55b447c2e0a989;hb=cb64cb72d5fff242cd90c13e4d8b9ca43e6e8837;hp=09e53620e343512a3e5a28c21433b03c3343d429;hpb=90f27333118e72b35068d7d7ac29e7b14f27aa3b;p=culture.git

diff --git a/reasoning.py b/reasoning.py
index 09e5362..c545e97 100755
--- a/reasoning.py
+++ b/reasoning.py
@@ -293,7 +293,7 @@ class Reasoning(problem.Problem):
                 X[i1:i2, j1:j2] = c[n]
                 f_X[i1:i2, j1:j2] = c[n if n > 0 else -1]
 
-    def task_move(self, A, f_A, B, f_B):
+    def task_translate(self, A, f_A, B, f_B):
         di, dj = torch.randint(3, (2,)) - 1
         nb_rec = 3
         c = torch.randperm(len(self.colors) - 1)[:nb_rec] + 1
@@ -406,16 +406,168 @@ class Reasoning(problem.Problem):
                 if n < nb_rec - 1:
                     f_X[i1, j1] = c[-1]
 
+    def task_count(self, A, f_A, B, f_B):
+        N = torch.randint(4, (1,)) + 2
+        c = torch.randperm(len(self.colors) - 1)[:N] + 1
+
+        for X, f_X in [(A, f_A), (B, f_B)]:
+
+            def contact(i, j, q):
+                nq, nq_diag = 0, 0
+                no = 0
+
+                for ii, jj in [
+                    (i - 1, j - 1),
+                    (i - 1, j),
+                    (i - 1, j + 1),
+                    (i, j - 1),
+                    (i, j + 1),
+                    (i + 1, j - 1),
+                    (i + 1, j),
+                    (i + 1, j + 1),
+                ]:
+                    if ii >= 0 and ii < self.height and jj >= 0 and jj < self.width:
+                        if X[ii, jj] != 0 and X[ii, jj] != q:
+                            no += 1
+
+                for ii, jj in [
+                    (i - 1, j - 1),
+                    (i - 1, j + 1),
+                    (i + 1, j - 1),
+                    (i + 1, j + 1),
+                ]:
+                    if ii >= 0 and ii < self.height and jj >= 0 and jj < self.width:
+                        if X[ii, jj] == q and X[i, jj] != q and X[ii, j] != q:
+                            nq_diag += 1
+
+                for ii, jj in [(i - 1, j), (i, j - 1), (i, j + 1), (i + 1, j)]:
+                    if ii >= 0 and ii < self.height and jj >= 0 and jj < self.width:
+                        if X[ii, jj] == q:
+                            nq += 1
+
+                return no, nq, nq_diag
+
+            nb = torch.zeros(N, dtype=torch.int64)
+            q = torch.randint(N, (self.height * self.width,))
+            k = torch.randperm(self.height * self.width)
+            for p in range(self.height * self.width):
+                i, j = k[p] % self.height, k[p] // self.height
+                no, nq, nq_diag = contact(i, j, c[q[p]])
+                if no == 0 and nq_diag == 0:
+                    if nq == 0:
+                        if nb[q[p]] < self.width:
+                            X[i, j] = c[q[p]]
+                            nb[q[p]] += 1
+                    if nq == 1:
+                        X[i, j] = c[q[p]]
+
+            for n in range(N):
+                for j in range(nb[n]):
+                    f_X[n, j] = c[n]
+
+    def task_trajectory(self, A, f_A, B, f_B):
+        c = torch.randperm(len(self.colors) - 1)[:2] + 1
+        for X, f_X in [(A, f_A), (B, f_B)]:
+            while True:
+                di, dj = torch.randint(7, (2,)) - 3
+                i, j = torch.randint(self.height, (1,)), torch.randint(self.width, (1,))
+                if (
+                    abs(di) + abs(dj) > 0
+                    and i + 2 * di >= 0
+                    and i + 2 * di < self.height
+                    and j + 2 * dj >= 0
+                    and j + 2 * dj < self.width
+                ):
+                    break
+
+            k = 0
+            while (
+                i + k * di >= 0
+                and i + k * di < self.height
+                and j + k * dj >= 0
+                and j + k * dj < self.width
+            ):
+                if k < 2:
+                    X[i + k * di, j + k * dj] = c[k]
+                f_X[i + k * di, j + k * dj] = c[min(k, 1)]
+                k += 1
+
+    def task_bounce(self, A, f_A, B, f_B):
+        c = torch.randperm(len(self.colors) - 1)[:3] + 1
+        for X, f_X in [(A, f_A), (B, f_B)]:
+
+            def free(i, j):
+                return (
+                    i >= 0
+                    and i < self.height
+                    and j >= 0
+                    and j < self.width
+                    and f_X[i, j] == 0
+                )
+
+            while True:
+                f_X[...] = 0
+                X[...] = 0
+
+                for _ in range((self.height * self.width) // 10):
+                    i, j = torch.randint(self.height, (1,)), torch.randint(
+                        self.width, (1,)
+                    )
+                    X[i, j] = c[0]
+                    f_X[i, j] = c[0]
+
+                while True:
+                    di, dj = torch.randint(7, (2,)) - 3
+                    if abs(di) + abs(dj) == 1:
+                        break
+
+                i, j = torch.randint(self.height, (1,)), torch.randint(self.width, (1,))
+
+                X[i, j] = c[1]
+                f_X[i, j] = c[1]
+                l = 0
+
+                while True:
+                    l += 1
+                    if free(i + di, j + dj):
+                        pass
+                    elif free(i - dj, j + di):
+                        di, dj = -dj, di
+                        if free(i + dj, j - di):
+                            if torch.rand(1) < 0.5:
+                                di, dj = -di, -dj
+                    elif free(i + dj, j - di):
+                        di, dj = dj, -di
+                    else:
+                        break
+
+                    i, j = i + di, j + dj
+                    f_X[i, j] = c[2]
+                    if l <= 1:
+                        X[i, j] = c[2]
+
+                    if l >= self.width:
+                        break
+
+                f_X[i, j] = c[1]
+                X[i, j] = c[1]
+
+                if l > 3:
+                    break
+
     ######################################################################
 
     def generate_prompts_and_answers(self, nb, device="cpu"):
         tasks = [
             self.task_replace_color,
-            self.task_move,
+            self.task_translate,
             self.task_grow,
             self.task_color_grow,
             self.task_frame,
             self.task_detect,
+            self.task_count,
+            self.task_trajectory,
+            self.task_bounce,
         ]
         prompts = torch.zeros(nb, self.height, self.width * 3, dtype=torch.int64)
         answers = torch.zeros(nb, self.height, self.width, dtype=torch.int64)
@@ -467,8 +619,8 @@ if __name__ == "__main__":
     delay = time.perf_counter() - start_time
     print(f"{prompts.size(0)/delay:02f} seq/s")
 
-    # predicted_prompts = torch.rand(prompts.size(0)) < 0.5
-    # predicted_answers = torch.logical_not(predicted_prompts)
+    predicted_prompts = torch.rand(prompts.size(0)) < 0.5
+    predicted_answers = torch.logical_not(predicted_prompts)
 
     reasoning.save_quizzes(
         "/tmp",
@@ -476,5 +628,6 @@ if __name__ == "__main__":
         prompts[:64],
         answers[:64],
         # You can add a bool to put a frame around the predicted parts
-        # predicted_prompts, predicted_answers
+        # predicted_prompts[:64],
+        # predicted_answers[:64],
     )