Update.
authorFrançois Fleuret <francois@fleuret.org>
Mon, 19 Feb 2024 15:47:27 +0000 (16:47 +0100)
committerFrançois Fleuret <francois@fleuret.org>
Mon, 19 Feb 2024 15:47:27 +0000 (16:47 +0100)
main.py
tasks.py

diff --git a/main.py b/main.py
index 9198edc..958dfc7 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -413,6 +413,7 @@ if args.task == "file":
         nb_train_samples=args.nb_train_samples,
         nb_test_samples=args.nb_test_samples,
         batch_size=args.batch_size,
+        shuffle=True,
         device=device,
     )
     args.max_percents_of_test_in_train = 0
index e5d3a7e..d21e264 100755 (executable)
--- a/tasks.py
+++ b/tasks.py
@@ -71,7 +71,7 @@ class Task:
 
 
 class TaskFromFile(Task):
-    def tensorize(self, pairs):
+    def tensorize(self, pairs, shuffle):
         len_max = max([len(x[0]) for x in pairs])
 
         input = torch.cat(
@@ -98,6 +98,12 @@ class TaskFromFile(Task):
             0,
         ).to("cpu")
 
+        if shuffle:
+            print("SHUFFLING!")
+            i = torch.randperm(input.size(0))
+            input = input[i].contiguous()
+            pred_mask = pred_mask[i].contiguous()
+
         return input, pred_mask
 
     # trim all the tensors in the tuple z to remove as much token from
@@ -122,6 +128,7 @@ class TaskFromFile(Task):
         nb_train_samples,
         nb_test_samples,
         batch_size,
+        shuffle=False,
         device=torch.device("cpu"),
     ):
         self.batch_size = batch_size
@@ -156,8 +163,12 @@ class TaskFromFile(Task):
         self.char2id = dict([(c, n) for n, c in enumerate(symbols)])
         self.id2char = dict([(n, c) for c, n in self.char2id.items()])
 
-        self.train_input, self.train_pred_masks = self.tensorize(train_pairs)
-        self.test_input, self.test_pred_masks = self.tensorize(test_pairs)
+        self.train_input, self.train_pred_masks = self.tensorize(
+            train_pairs, shuffle=shuffle
+        )
+        self.test_input, self.test_pred_masks = self.tensorize(
+            test_pairs, shuffle=shuffle
+        )
 
     def batches(self, split="train", nb_to_use=-1, desc=None):
         assert split in {"train", "test"}