X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;ds=sidebyside;f=tasks.py;h=0f44760093b1dbe9ccb9835663b4e1772d9a9b4b;hb=439c597d409c344283f8996f042daf79d3f24de2;hp=75cd35ed6c2e2c280fa4ad2ea07e7adb86439bb8;hpb=0c47d4d8ef8c4938f4765af816349cf30da14cb1;p=picoclvr.git

diff --git a/tasks.py b/tasks.py
index 75cd35e..0f44760 100755
--- a/tasks.py
+++ b/tasks.py
@@ -1052,6 +1052,11 @@ class RPL(Task):
         nb_train_samples,
         nb_test_samples,
         batch_size,
+        nb_starting_values=3,
+        max_input=9,
+        prog_len=6,
+        nb_runs=5,
+        logger=None,
         device=torch.device("cpu"),
     ):
         super().__init__()
@@ -1060,11 +1065,23 @@ class RPL(Task):
         self.device = device
 
         train_sequences = [
-            rpl.generate()
+            rpl.generate(
+                nb_starting_values=nb_starting_values,
+                max_input=max_input,
+                prog_len=prog_len,
+                nb_runs=nb_runs,
+            )
             for _ in tqdm.tqdm(range(nb_train_samples), desc="train-data")
         ]
+
         test_sequences = [
-            rpl.generate() for _ in tqdm.tqdm(range(nb_test_samples), desc="test-data")
+            rpl.generate(
+                nb_starting_values=nb_starting_values,
+                max_input=max_input,
+                prog_len=prog_len,
+                nb_runs=nb_runs,
+            )
+            for _ in tqdm.tqdm(range(nb_test_samples), desc="test-data")
         ]
 
         symbols = list(
@@ -1083,6 +1100,13 @@ class RPL(Task):
         self.train_input = self.tensorize(train_sequences)
         self.test_input = self.tensorize(test_sequences)
 
+        if logger is not None:
+            for x in self.train_input[:10]:
+                end = (x != self.t_nul).nonzero().max().item() + 1
+                seq = [self.id2token[i.item()] for i in x[:end]]
+                s = " ".join(seq)
+                logger(f"example_seq {s}")
+
         self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
 
     def batches(self, split="train", nb_to_use=-1, desc=None):
@@ -1131,14 +1155,15 @@ class RPL(Task):
                     _, _, gt_prog, _ = rpl.compute_nb_errors(gt_seq)
                     gt_prog = " ".join([str(x) for x in gt_prog])
                     prog = " ".join([str(x) for x in prog])
-                    logger(f"GROUND-TRUTH PROG [{gt_prog}] PREDICTED PROG [{prog}]")
+                    comment = "*" if nb_errors == 0 else "-"
+                    logger(f"{comment} PROG [{gt_prog}] PREDICTED [{prog}]")
                     for start_stack, target_stack, result_stack, correct in stacks:
-                        comment = " CORRECT" if correct else ""
+                        comment = "*" if correct else "-"
                         start_stack = " ".join([str(x) for x in start_stack])
                         target_stack = " ".join([str(x) for x in target_stack])
                         result_stack = " ".join([str(x) for x in result_stack])
                         logger(
-                            f"  [{start_stack}] -> [{result_stack}] TARGET [{target_stack}]{comment}"
+                            f"  {comment} [{start_stack}] -> [{target_stack}] PREDICTED [{result_stack}]"
                         )
                     nb_to_log -= 1