Update.
authorFrançois Fleuret <francois@fleuret.org>
Sat, 15 Jul 2023 08:21:09 +0000 (10:21 +0200)
committerFrançois Fleuret <francois@fleuret.org>
Sat, 15 Jul 2023 08:21:09 +0000 (10:21 +0200)
main.py
tasks.py
world.py

diff --git a/main.py b/main.py
index c763016..58e8046 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -133,6 +133,11 @@ parser.add_argument("--expr_result_max", type=int, default=99)
 
 parser.add_argument("--expr_input_file", type=str, default=None)
 
+##############################
+# World options
+
+parser.add_argument("--world_vqae_nb_epochs", type=int, default=10)
+
 ######################################################################
 
 args = parser.parse_args()
@@ -328,6 +333,7 @@ elif args.task == "world":
         nb_train_samples=args.nb_train_samples,
         nb_test_samples=args.nb_test_samples,
         batch_size=args.batch_size,
+        vqae_nb_epochs=args.world_vqae_nb_epochs,
         device=device,
     )
 
index 15d97b8..96d0621 100755 (executable)
--- a/tasks.py
+++ b/tasks.py
@@ -20,6 +20,8 @@ def masked_inplace_autoregression(
     progress_bar_desc="autoregression",
     device=torch.device("cpu"),
 ):
+    assert input.size() == ar_mask.size()
+
     batches = zip(input.split(batch_size), ar_mask.split(batch_size))
 
     if progress_bar_desc is not None:
@@ -27,7 +29,7 @@ def masked_inplace_autoregression(
             batches,
             dynamic_ncols=True,
             desc=progress_bar_desc,
-            total=input.size(0) // batch_size,
+            #total=input.size(0) // batch_size,
         )
 
     with torch.autograd.no_grad():
@@ -944,6 +946,7 @@ class Expr(Task):
 
 
 ######################################################################
+
 import world
 
 
@@ -953,15 +956,16 @@ class World(Task):
         nb_train_samples,
         nb_test_samples,
         batch_size,
+        vqae_nb_epochs,
         device=torch.device("cpu"),
     ):
         self.batch_size = batch_size
         self.device = device
 
         (
-            self.train_input,
+            train_frames,
             self.train_actions,
-            self.test_input,
+            test_frames,
             self.test_actions,
             self.frame2seq,
             self.seq2frame,
@@ -970,9 +974,15 @@ class World(Task):
             nb_test_samples,
             mode="first_last",
             nb_steps=30,
-            nb_epochs=2,
+            nb_epochs=vqae_nb_epochs,
+            device=device,
         )
 
+        self.train_input = self.frame2seq(train_frames)
+        self.train_input = self.train_input.reshape(self.train_input.size(0) // 2, -1)
+        self.test_input = self.frame2seq(test_frames)
+        self.test_input = self.test_input.reshape(self.test_input.size(0) // 2, -1)
+
         self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
 
     def batches(self, split="train", nb_to_use=-1, desc=None):
@@ -993,7 +1003,34 @@ class World(Task):
     def produce_results(
         self, n_epoch, model, result_dir, logger, deterministic_synthesis
     ):
-        pass
+        l = self.train_input.size(1)
+        k = torch.arange(l, device=self.device)[None, :]
+        result = self.test_input[:64].clone()
+
+        ar_mask = (k >= l // 2).long().expand_as(result)
+        result *= 1 - ar_mask
+
+        masked_inplace_autoregression(
+            model,
+            self.batch_size,
+            result,
+            ar_mask,
+            deterministic_synthesis,
+            device=self.device,
+        )
+
+        result = result.reshape(result.size(0) * 2, -1)
+
+        frames = self.seq2frame(result)
+        image_name = os.path.join(result_dir, f"world_result_{n_epoch:04d}.png")
+        torchvision.utils.save_image(
+            frames.float() / (world.Box.nb_rgb_levels - 1),
+            image_name,
+            nrow=8,
+            padding=1,
+            pad_value=0.0,
+        )
+        logger(f"wrote {image_name}")
 
 
 ######################################################################
index c3eb101..5c21fad 100755 (executable)
--- a/world.py
+++ b/world.py
@@ -65,7 +65,7 @@ class SignSTE(nn.Module):
 def train_encoder(
     train_input,
     test_input,
-    depth=3,
+    depth=2,
     dim_hidden=48,
     nb_bits_per_token=8,
     lr_start=1e-3,
@@ -331,7 +331,12 @@ def generate_episodes(nb, steps):
 
 
 def create_data_and_processors(
-    nb_train_samples, nb_test_samples, mode, nb_steps, nb_epochs=10
+    nb_train_samples,
+    nb_test_samples,
+    mode,
+    nb_steps,
+    nb_epochs=10,
+    device=torch.device("cpu"),
 ):
     assert mode in ["first_last"]
 
@@ -339,10 +344,12 @@ def create_data_and_processors(
         steps = [True] + [False] * (nb_steps + 1) + [True]
 
     train_input, train_actions = generate_episodes(nb_train_samples, steps)
+    train_input, train_actions = train_input.to(device), train_actions.to(device)
     test_input, test_actions = generate_episodes(nb_test_samples, steps)
+    test_input, test_actions = test_input.to(device), test_actions.to(device)
 
     encoder, quantizer, decoder = train_encoder(
-        train_input, test_input, nb_epochs=nb_epochs
+        train_input, test_input, nb_epochs=nb_epochs, device=device
     )
     encoder.train(False)
     quantizer.train(False)