def save_image(self, input, result_dir, filename, logger):
img = world.sample2img(input.to("cpu"), self.height, self.width)
image_name = os.path.join(result_dir, filename)
- torchvision.utils.save_image(img.float() / 255.0, image_name, nrow=8, padding=2)
+ torchvision.utils.save_image(img.float() / 255.0, image_name, nrow=6, padding=4)
logger(f"wrote {image_name}")
def make_ar_mask(self, input):
self.height = 6
self.width = 8
- self.train_input = world.generate(
+ self.train_input = world.generate_seq(
nb_train_samples, height=self.height, width=self.width
).to(device)
- self.test_input = world.generate(
+ self.test_input = world.generate_seq(
nb_test_samples, height=self.height, width=self.width
).to(device)
+ # print()
+ # for a in world.seq2str(self.train_input):
+ # print(a)
+ # for a in world.seq2str(self.test_input):
+ # print(a)
+ # exit(0)
+
self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
self.train_quizzes = []
if result_dir is not None:
self.save_image(
- self.train_input[:96], result_dir, f"world_train.png", logger
+ self.train_input[:72], result_dir, f"world_train.png", logger
)
def batches(self, split="train", desc=None):
)
self.save_image(
- result[:96],
+ result[:72],
result_dir,
f"world_prediction_{n_epoch:04d}_{model.id:02d}.png",
logger,
# Check how many of the other models can solve them in both
# directions
- nb_correct = 0
+ nb_correct = []
for m in other_models:
result = quizzes.clone()
(reverse_quizzes == reverse_result).long().min(dim=-1).values
)
- nb_correct += correct * reverse_correct
+ nb_correct.append((correct * reverse_correct)[None, :])
+
+ nb_correct = torch.cat(nb_correct, dim=0)
+
+ filename = os.path.join(result_dir, "correct_{n_epoch:04d}.dat")
+ with open(filename, "w") as f:
+ for k in nb_correct:
+ f.write(f"{k}\n")
- return quizzes, nb_correct
+ return quizzes, nb_correct.sum(dim=0)