From: Francois Fleuret <francois@fleuret.org>
Date: Wed, 27 Jul 2022 14:42:28 +0000 (+0200)
Subject: OCDC
X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=commitdiff_plain;h=1ad9ea3cca4489b07bad8521966382f66a493eea;p=mygpt.git

OCDC
---

diff --git a/main.py b/main.py
index 339d185..5f3e8cf 100755
--- a/main.py
+++ b/main.py
@@ -156,7 +156,7 @@ import picoclvr
 
 class TaskPicoCLVR(Task):
 
-    def descr2tensor(self, descr):
+    def tensorize(self, descr):
         t = [ [ self.token2id[u] for u in s ] for s in descr ]
         return torch.tensor(t, device = self.device)
 
@@ -173,8 +173,8 @@ class TaskPicoCLVR(Task):
 
             descr = [ s.strip().split(' ') for s in descr ]
             l = max([ len(s) for s in descr ])
-            #descr = [ [ '<unk>' ] * (l - len(s)) + s for s in descr ]
-            descr = [ s + [ '<unk>' ] * (l - len(s)) for s in descr ]
+            #descr = [ [ '<nul>' ] * (l - len(s)) + s for s in descr ]
+            descr = [ s + [ '<nul>' ] * (l - len(s)) for s in descr ]
 
             return descr
 
@@ -196,8 +196,8 @@ class TaskPicoCLVR(Task):
         self.id2token = dict([ (n, t) for n, t in enumerate(tokens) ])
 
         # Tokenize the train and test sets
-        self.train_input = descr2tensor(self.train_descr)
-        self.test_input = descr2tensor(self.test_descr)
+        self.train_input = self.tensorize(self.train_descr)
+        self.test_input = self.tensorize(self.test_descr)
 
     def batches(self, split = 'train'):
         assert split in { 'train', 'test' }
@@ -208,14 +208,6 @@ class TaskPicoCLVR(Task):
     def vocabulary_size(self):
         return len(self.token2id)
 
-    def generate(self, primer_descr, model, nb_tokens):
-        results = autoregression(
-            model, self.batch_size,
-            nb_samples = 1, nb_tokens = nb_tokens, primer = descr2tensor(primer_descr),
-            device = self.device
-        )
-        return ' '.join([ self.id2token[t.item()] for t in results.flatten() ])
-
     def produce_results(self, n_epoch, model):
         nb_tokens = self.height * self.width + 3
         result_descr = [ ]
@@ -229,10 +221,20 @@ class TaskPicoCLVR(Task):
         ]:
 
             for k in range(nb_per_primer):
-                result_descr.append(self.generate(primer_descr, model, nb_tokens))
+                results = autoregression(
+                    model, self.batch_size,
+                    nb_samples = 1, nb_tokens = nb_tokens,
+                    primer = self.tensorize(primer_descr),
+                    device = self.device
+                )
+                r = ' '.join([ self.id2token[t.item()] for t in results.flatten() ])
+                result_descr.append(r)
+
+        img = [
+            picoclvr.descr2img(d, height = self.height, width = self.width)
+            for d in result_descr
+        ]
 
-        img = [ picoclvr.descr2img(d, height = self.height, width = self.width)
-                for d in result_descr ]
         img = torch.cat(img, 0)
         image_name = f'result_picoclvr_{n_epoch:04d}.png'
         torchvision.utils.save_image(
@@ -276,7 +278,7 @@ class TaskWiki103(Task):
 
         self.vocab = torchtext.vocab.build_vocab_from_iterator(
             yield_tokens(),
-            specials = [ '<unk>', '<non>' ],
+            specials = [ '<unk>', '<nul>' ],
             min_freq = self.min_freq
         )
 
@@ -284,7 +286,7 @@ class TaskWiki103(Task):
 
     def tensorize(self, s):
         a = max(len(x) for x in s)
-        return torch.tensor([ self.vocab(x + [ '<non>' ] * (a - len(x))) for x in s ])
+        return torch.tensor([ self.vocab(x + [ '<nul>' ] * (a - len(x))) for x in s ])
 
     def yield_batches(self, ds):
         s = [ ]
@@ -342,7 +344,7 @@ class TaskWiki103(Task):
                      else:
                          t_next = logits.argmax()
                      t_generated.append(self.vocab.lookup_token(t_next))
-                     if t_generated[-1] == '<non>': break
+                     if t_generated[-1] == '<nul>': break
 
                  s = ' '.join(t_generated)