Update.
authorFrançois Fleuret <francois@fleuret.org>
Thu, 6 Jul 2023 09:30:42 +0000 (11:30 +0200)
committerFrançois Fleuret <francois@fleuret.org>
Thu, 6 Jul 2023 09:30:42 +0000 (11:30 +0200)
main.py
maze.py
mygpt.py
tasks.py [new file with mode: 0755]

diff --git a/main.py b/main.py
index 9dee679..5b49468 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -14,7 +14,7 @@ import torch, torchvision
 from torch import nn
 from torch.nn import functional as F
 
-import mygpt, tensorstack
+import mygpt, tasks, tensorstack
 
 ######################################################################
 
@@ -92,11 +92,11 @@ parser.add_argument("--picocvlr_prune_properties", type=str, default="none")
 ##############################
 # Maze options
 
-parser.add_argument("--maze_height", type=int, default=13)
+parser.add_argument("--maze_height", type=int, default=23)
 
-parser.add_argument("--maze_width", type=int, default=21)
+parser.add_argument("--maze_width", type=int, default=39)
 
-parser.add_argument("--maze_nb_walls", type=int, default=15)
+parser.add_argument("--maze_nb_walls", type=int, default=45)
 
 ##############################
 # Snake options
@@ -114,11 +114,11 @@ parser.add_argument("--snake_length", type=int, default=200)
 
 parser.add_argument("--stack_nb_steps", type=int, default=100)
 
-parser.add_argument("--stack_nb_stacks", type=int, default=1)
+parser.add_argument("--stack_nb_stacks", type=int, default=3)
 
 parser.add_argument("--stack_nb_digits", type=int, default=3)
 
-parser.add_argument("--stack_fraction_values_for_train", type=float, default=None)
+parser.add_argument("--stack_fraction_values_for_train", type=float, default=0.75)
 
 ##############################
 # Expr options
@@ -153,7 +153,7 @@ default_args = {
     },
     "maze": {
         "nb_epochs": 25,
-        "batch_size": 25,
+        "batch_size": 5,
         "nb_train_samples": 250000,
         "nb_test_samples": 10000,
     },
@@ -218,969 +218,6 @@ def log_string(s):
 for n in vars(args):
     log_string(f"args.{n} {getattr(args, n)}")
 
-######################################################################
-
-
-# ra_mask is boolean, with 1s on the values to generate
-
-
-def masked_inplace_autoregression(
-    model,
-    batch_size,
-    input,
-    ar_mask,
-    forbidden_tokens=None,
-    progress_bar_desc="autoregression",
-    device=torch.device("cpu"),
-):
-    batches = zip(input.split(batch_size), ar_mask.split(batch_size))
-
-    if progress_bar_desc is not None:
-        batches = tqdm.tqdm(
-            batches,
-            dynamic_ncols=True,
-            desc=progress_bar_desc,
-            total=input.size(0) // batch_size,
-        )
-
-    for input, ar_mask in batches:
-        i = (ar_mask.sum(0) > 0).nonzero()
-        if i.min() > 0:
-            model(
-                mygpt.BracketedSequence(input, 0, i.min())
-            )  # Needed to initialize the model's cache
-        for s in range(i.min(), i.max() + 1):
-            output = model(mygpt.BracketedSequence(input, s, 1)).x
-            logits = output[:, s]
-            if forbidden_tokens is not None:
-                logits = logits.masked_fill(forbidden_tokens, float("-inf"))
-            if args.deterministic_synthesis:
-                t_next = logits.argmax(1)
-            else:
-                dist = torch.distributions.categorical.Categorical(logits=logits)
-                t_next = dist.sample()
-            input[:, s] = ar_mask[:, s] * t_next + (1 - ar_mask[:, s]) * input[:, s]
-
-
-######################################################################
-
-
-class Task:
-    def batches(self, split="train"):
-        pass
-
-    def vocabulary_size(self):
-        pass
-
-    def produce_results(self, n_epoch, model):
-        pass
-
-
-######################################################################
-
-import picoclvr
-
-
-class TaskPicoCLVR(Task):
-    # Make a tensor from a list of strings
-    def tensorize(self, descr):
-        token_descr = [s.strip().split(" ") for s in descr]
-        l = max([len(s) for s in token_descr])
-        token_descr = [s + ["<nul>"] * (l - len(s)) for s in token_descr]
-        id_descr = [[self.token2id[u] for u in s] for s in token_descr]
-        return torch.tensor(id_descr, device=self.device)
-
-    # Make a list of strings from a tensor
-    def detensorize(self, x):
-        return [" ".join([self.id2token[t.item()] for t in r]) for r in x]
-
-    # trim all the tensors in the tuple z to remove as much token from
-    # left and right in the first tensor. If z is a tuple, all its
-    # elements are trimed according to the triming for the first
-    def trim(self, z, token="<nul>"):
-        n = self.token2id[token]
-        if type(z) == tuple:
-            x = z[0]
-            i = (1 - (F.pad(x, (1, 1), value=n) == n).min(0).values.long()).cumsum(0)
-            a, b = (i == 0).nonzero().max(), (i == i.max()).nonzero().min()
-            return tuple([t[:, a:b] for t in z])
-        else:
-            i = (1 - (F.pad(z, (1, 1), value=n) == n).min(0).values.long()).cumsum(0)
-            a, b = (i == 0).nonzero().max(), (i == i.max()).nonzero().min()
-            return z[:, a:b]
-
-    ######################
-    # Not the cleanest part of the code
-
-    # Extract the last image of each sequence, from the last <img>
-    # included, and set to <nul> all the tokens from the beginning of
-    # that image to the end
-    def excise_last_image(self, input):
-        t_img, t_nul = self.token2id["<img>"], self.token2id["<nul>"]
-        nb_img_tokens = self.height * self.width + 1
-
-        input = input.clone()
-        t = (input == t_img).long()
-        tail_masks = (t.cumsum(dim=1) == t.sum(dim=1, keepdim=True)).long()
-        i = (t * tail_masks).nonzero(as_tuple=True)
-        j = (
-            i[0][:, None],
-            i[1][:, None] + torch.arange(nb_img_tokens, device=input.device)[None, :],
-        )
-        images = self.trim(input[j])
-        input[j] = t_nul
-        loss_masks = 1 - tail_masks
-        input, loss_masks = self.trim((input, loss_masks))
-        return input, loss_masks, images
-
-    def add_true_image(self, input, images, loss_masks):
-        t_nul = self.token2id["<nul>"]
-        nb_img_tokens = self.height * self.width + 1
-        input = F.pad(input, (0, nb_img_tokens), value=t_nul)
-        loss_masks = F.pad(loss_masks, (0, nb_img_tokens), value=0)
-        t = (input == t_nul).long()
-        i = (t.cumsum(dim=1) == 1).nonzero(as_tuple=True)
-        j = (
-            i[0][:, None],
-            i[1][:, None] + torch.arange(nb_img_tokens, device=input.device)[None, :],
-        )
-        input[j] = images
-        loss_masks[j] = 1
-        input, loss_masks = self.trim((input, loss_masks))
-        return input, loss_masks
-
-    def add_generated_image(self, input, loss_masks, model):
-        t_img, t_nul = self.token2id["<img>"], self.token2id["<nul>"]
-        nb_img_tokens = self.height * self.width + 1
-
-        input = F.pad(input, (0, nb_img_tokens), value=t_nul)
-        loss_masks = F.pad(loss_masks, (0, nb_img_tokens), value=0)
-        t = (input == t_nul).long()
-        i = (t.cumsum(dim=1) == 1).nonzero(as_tuple=True)
-        input[i] = t_img
-
-        j = (
-            i[0][:, None],
-            i[1][:, None]
-            + 1
-            + torch.arange(nb_img_tokens - 1, device=input.device)[None, :],
-        )
-        ar_masks = input.new_zeros(input.size(), dtype=torch.int64)
-        ar_masks[j] = 1
-        forbidden_tokens = (
-            torch.arange(self.vocabulary_size(), device=input.device) == t_nul
-        )
-        with torch.autograd.no_grad():
-            t = model.training
-            model.eval()
-            masked_inplace_autoregression(
-                model,
-                self.batch_size,
-                input,
-                ar_masks,
-                forbidden_tokens,
-                progress_bar_desc=None,
-                device=self.device,
-            )
-            model.train(t)
-
-        input, loss_masks = self.trim((input, loss_masks))
-
-        return input, loss_masks
-
-    ######################
-
-    def __init__(
-        self,
-        nb_train_samples,
-        nb_test_samples,
-        batch_size,
-        height,
-        width,
-        nb_colors=5,
-        device=torch.device("cpu"),
-        pruner_train=None,
-        pruner_eval=None,
-    ):
-        def generate_descr(nb, cache_suffix, pruner):
-            return picoclvr.generate(
-                nb,
-                height=self.height,
-                width=self.width,
-                nb_colors=nb_colors,
-                pruner=pruner,
-            )
-
-        self.height = height
-        self.width = width
-        self.batch_size = batch_size
-        self.device = device
-        self.pruner_train = pruner_train
-        self.pruner_eval = pruner_eval
-
-        param = {
-            "nb_train_samples": nb_train_samples,
-            "nb_test_samples": nb_test_samples,
-            "height": height,
-            "width": width,
-            "nb_colors": nb_colors,
-            "batch_size": batch_size,
-            "rng_state": list(torch.get_rng_state()),
-        }
-
-        log_string(
-            f"generating {nb_train_samples+nb_test_samples} samples (can take some time)"
-        )
-        self.train_descr = generate_descr(
-            nb_train_samples, "train", pruner=self.pruner_train
-        )
-        self.test_descr = generate_descr(nb_test_samples, "test", pruner=None)
-
-        # Build the tokenizer
-        tokens = {"<nul>", "<img>"}
-        for d in [self.train_descr, self.test_descr]:
-            for s in d:
-                for t in s.strip().split(" "):
-                    tokens.add(t)
-        # make this set a sorted list to get the same tensors given
-        # the same descr
-        tokens = list(tokens)
-        tokens.sort()
-        self.token2id = dict([(t, n) for n, t in enumerate(tokens)])
-        self.id2token = dict([(n, t) for n, t in enumerate(tokens)])
-
-        # Tokenize the train and test sets
-        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"}
-        input = self.train_input if split == "train" else self.test_input
-        for batch in tqdm.tqdm(
-            input.split(self.batch_size), dynamic_ncols=True, desc=f"epoch-{split}"
-        ):
-            yield self.trim(batch)
-
-    def vocabulary_size(self):
-        return len(self.token2id)
-
-    def compute_missing_properties(self, n_epoch, model, pruner=None):
-        acc_nb_requested_properties = []
-        acc_nb_missing_properties = []
-        acc_nb_results = 0
-
-        for input in tqdm.tqdm(
-            self.test_input.split(self.batch_size),
-            dynamic_ncols=True,
-            desc=f"test-properties",
-        ):
-            tape, loss_masks, _ = self.excise_last_image(input)
-            tape, loss_masks = self.add_generated_image(tape, loss_masks, model)
-            result_descr = self.detensorize(tape)
-            np = picoclvr.nb_properties(
-                result_descr,
-                height=self.height,
-                width=self.width,
-                pruner=pruner,
-            )
-            nb_requested_properties, _, nb_missing_properties = zip(*np)
-            acc_nb_requested_properties += nb_requested_properties
-            acc_nb_missing_properties += nb_missing_properties
-            acc_nb_results += len(result_descr)
-
-        nb_requested_properties = sum(acc_nb_requested_properties)
-        nb_missing_properties = sum(acc_nb_missing_properties)
-
-        prefix = "" if pruner is None else "pruned_"
-        log_string(f"nb_{prefix}samples {n_epoch} {acc_nb_results}")
-        log_string(
-            f"property_{prefix}nb {n_epoch} requested {sum(acc_nb_requested_properties)} missing {sum(acc_nb_missing_properties)}"
-        )
-        log_string(
-            f"property_{prefix}miss {n_epoch} {100*nb_missing_properties/nb_requested_properties:.02f}%"
-        )
-
-    ######################################################################
-
-    def produce_results(self, n_epoch, model):
-        self.compute_missing_properties(n_epoch, model)
-
-        if self.pruner_eval is not None:
-            self.compute_missing_properties(n_epoch, model, self.pruner_eval)
-
-        nb_tokens_to_generate = self.height * self.width + 3
-        result_descr = []
-        nb_per_primer = 8
-        primer = []
-
-        for primer_descr in [
-            "red above green <sep> green top <sep> blue right of red",
-            "there is red <sep> there is yellow <sep> there is blue",
-            "red below yellow <sep> yellow below green <sep> green below blue <sep> red right <sep> yellow left <sep> green right <sep> blue left",
-            "green bottom <sep> yellow bottom <sep> green left of blue <sep> yellow right of blue <sep> blue top",
-        ]:
-            primer += [primer_descr] * nb_per_primer
-
-        tape = self.tensorize(primer)
-        loss_masks = 1 - (tape == self.token2id["<nul>"]).long()
-        tape, loss_masks = self.add_generated_image(tape, loss_masks, model)
-        result_descr = self.detensorize(tape)
-
-        np = picoclvr.nb_properties(result_descr, height=self.height, width=self.width)
-
-        acc_nb_requested_properties, _, acc_nb_missing_properties = zip(*np)
-        acc_nb_results = len(result_descr)
-
-        nb_requested_properties = sum(acc_nb_requested_properties)
-        nb_missing_properties = sum(acc_nb_missing_properties)
-
-        prefix = "demo_"
-        log_string(f"nb_{prefix}samples {n_epoch} {acc_nb_results}")
-        log_string(
-            f"property_{prefix}nb {n_epoch} requested {sum(acc_nb_requested_properties)} missing {sum(acc_nb_missing_properties)}"
-        )
-        log_string(
-            f"property_{prefix}miss {n_epoch} {100*nb_missing_properties/nb_requested_properties:.02f}%"
-        )
-
-        img = picoclvr.descr2img(result_descr, height=self.height, width=self.width)
-
-        if img.dim() == 5:
-            if img.size(1) == 1:
-                img = F.pad(img.squeeze(1), pad=(1, 1, 1, 1), value=64)
-            else:
-                img = torch.cat(
-                    [
-                        torchvision.utils.make_grid(x, padding=1, pad_value=64)[None]
-                        for x in img
-                    ],
-                    0,
-                )
-
-        image_name = os.path.join(args.result_dir, f"picoclvr_result_{n_epoch:04d}.png")
-        torchvision.utils.save_image(
-            img / 255.0, image_name, nrow=nb_per_primer, padding=1, pad_value=0.0
-        )
-        log_string(f"wrote {image_name}")
-
-
-######################################################################
-
-
-class TaskMNIST(Task):
-    def __init__(self, batch_size, device=torch.device("cpu")):
-        self.device = device
-        self.batch_size = batch_size
-
-    def batches(self, split="train"):
-        assert split in {"train", "test"}
-        data_set = torchvision.datasets.MNIST(
-            root="./data", train=(split == "train"), download=True
-        )
-        data_input = data_set.data.view(-1, 28 * 28).long()
-        if args.nb_train_samples is not None:
-            data_input = data_input[: args.nb_train_samples]
-        for batch in tqdm.tqdm(
-            data_input.split(self.batch_size), desc=f"epoch-{split}"
-        ):
-            yield batch
-
-    def vocabulary_size(self):
-        return 256
-
-    def produce_results(self, n_epoch, model):
-        results = torch.empty(64, 28 * 28, device=self.device, dtype=torch.int64)
-        ar_mask = torch.full_like(results, 1)
-        masked_inplace_autoregression(
-            model, self.batch_size, results, ar_mask, device=self.device
-        )
-        image_name = os.path.join(args.result_dir, f"mnist_result_{n_epoch:04d}.png")
-        torchvision.utils.save_image(
-            1 - results.reshape(-1, 1, 28, 28) / 255.0,
-            image_name,
-            nrow=16,
-            pad_value=0.8,
-        )
-        log_string(f"wrote {image_name}")
-
-
-######################################################################
-
-import maze
-
-
-class TaskMaze(Task):
-    def map2seq(self, *m):
-        return torch.cat([x.flatten(1) for x in m], 1)
-
-    def seq2map(self, s):
-        s = s.reshape(s.size(0), -1, self.height, self.width)
-        return (s[:, k] for k in range(s.size(1)))
-
-    def __init__(
-        self,
-        nb_train_samples,
-        nb_test_samples,
-        batch_size,
-        height,
-        width,
-        nb_walls,
-        device=torch.device("cpu"),
-    ):
-        self.batch_size = batch_size
-        self.height = height
-        self.width = width
-        self.device = device
-
-        train_mazes, train_paths, _ = maze.create_maze_data(
-            nb_train_samples,
-            height=height,
-            width=width,
-            nb_walls=nb_walls,
-            progress_bar=lambda x: tqdm.tqdm(x, dynamic_ncols=True, desc=f"data-train"),
-        )
-        self.train_input = self.map2seq(train_mazes.to(device), train_paths.to(device))
-
-        test_mazes, test_paths, _ = maze.create_maze_data(
-            nb_test_samples,
-            height=height,
-            width=width,
-            nb_walls=nb_walls,
-            progress_bar=lambda x: tqdm.tqdm(x, dynamic_ncols=True, desc=f"data-test"),
-        )
-        self.test_input = self.map2seq(test_mazes.to(device), test_paths.to(device))
-
-        self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
-
-    def batches(self, split="train", nb_to_use=-1, desc=None):
-        assert split in {"train", "test"}
-        input = self.train_input if split == "train" else self.test_input
-        if nb_to_use > 0:
-            input = input[:nb_to_use]
-        if desc is None:
-            desc = f"epoch-{split}"
-        for batch in tqdm.tqdm(
-            input.split(self.batch_size), dynamic_ncols=True, desc=desc
-        ):
-            yield batch
-
-    def vocabulary_size(self):
-        return self.nb_codes
-
-    def compute_error(self, model, split="train", nb_to_use=-1):
-        nb_total, nb_correct = 0, 0
-        count = torch.zeros(
-            self.width * self.height,
-            self.width * self.height,
-            device=self.device,
-            dtype=torch.int64,
-        )
-        for input in tqdm.tqdm(
-            task.batches(split, nb_to_use),
-            dynamic_ncols=True,
-            desc=f"test-mazes",
-        ):
-            result = input.clone()
-            ar_mask = result.new_zeros(result.size())
-            ar_mask[:, self.height * self.width :] = 1
-            result *= 1 - ar_mask
-            masked_inplace_autoregression(
-                model,
-                self.batch_size,
-                result,
-                ar_mask,
-                progress_bar_desc=None,
-                device=self.device,
-            )
-            mazes, paths = self.seq2map(result)
-            path_correctness = maze.path_correctness(mazes, paths)
-            nb_correct += path_correctness.long().sum()
-            nb_total += mazes.size(0)
-
-            optimal_path_lengths = (
-                (input[:, self.height * self.width :] == maze.v_path).long().sum(1)
-            )
-            predicted_path_lengths = (
-                (result[:, self.height * self.width :] == maze.v_path).long().sum(1)
-            )
-            optimal_path_lengths = optimal_path_lengths[path_correctness]
-            predicted_path_lengths = predicted_path_lengths[path_correctness]
-            count[optimal_path_lengths, predicted_path_lengths] += 1
-
-        if count.max() == 0:
-            count = None
-        else:
-            count = count[
-                : count.sum(1).nonzero().max() + 1, : count.sum(0).nonzero().max() + 1
-            ]
-
-        return nb_total, nb_correct, count
-
-    def produce_results(self, n_epoch, model):
-        with torch.autograd.no_grad():
-            t = model.training
-            model.eval()
-
-            train_nb_total, train_nb_correct, count = self.compute_error(
-                model, "train", nb_to_use=1000
-            )
-            log_string(
-                f"accuracy_train {n_epoch} nb_total {train_nb_total} nb_correct {train_nb_correct} accuracy {(100.0*train_nb_correct)/train_nb_total:.02f}%"
-            )
-
-            test_nb_total, test_nb_correct, count = self.compute_error(
-                model, "test", nb_to_use=1000
-            )
-            log_string(
-                f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
-            )
-
-            if count is not None:
-                proportion_optimal = count.diagonal().sum().float() / count.sum()
-                log_string(f"proportion_optimal_test {proportion_optimal*100:.02f}%")
-                with open(
-                    os.path.join(args.result_dir, f"maze_result_{n_epoch:04d}.txt"), "w"
-                ) as f:
-                    for i in range(count.size(0)):
-                        for j in range(count.size(1)):
-                            eol = " " if j < count.size(1) - 1 else "\n"
-                            f.write(f"{count[i,j]}{eol}")
-
-            input = self.test_input[:48]
-            result = input.clone()
-            ar_mask = result.new_zeros(result.size())
-            ar_mask[:, self.height * self.width :] = 1
-            result *= 1 - ar_mask
-            masked_inplace_autoregression(
-                model, self.batch_size, result, ar_mask, device=self.device
-            )
-
-            mazes, paths = self.seq2map(input)
-            _, predicted_paths = self.seq2map(result)
-
-            filename = os.path.join(args.result_dir, f"maze_result_{n_epoch:04d}.png")
-            maze.save_image(
-                filename,
-                mazes=mazes,
-                target_paths=paths,
-                predicted_paths=predicted_paths,
-                path_correct=maze.path_correctness(mazes, predicted_paths),
-                path_optimal=maze.path_optimality(paths, predicted_paths),
-            )
-            log_string(f"wrote {filename}")
-
-            model.train(t)
-
-
-######################################################################
-
-
-import snake
-
-
-class TaskSnake(Task):
-    def __init__(
-        self,
-        nb_train_samples,
-        nb_test_samples,
-        batch_size,
-        height,
-        width,
-        nb_colors,
-        length,
-        prompt_length,
-        device=torch.device("cpu"),
-    ):
-        self.batch_size = batch_size
-        self.height = height
-        self.width = width
-        self.device = device
-        self.prompt_length = prompt_length
-
-        self.train_input, self.train_prior_visits, _, _ = snake.generate_sequences(
-            nb_train_samples,
-            height,
-            width,
-            nb_colors,
-            length,
-            prompt_length,
-            self.device,
-        )
-        self.test_input, self.test_prior_visits, _, _ = snake.generate_sequences(
-            nb_test_samples,
-            height,
-            width,
-            nb_colors,
-            length,
-            prompt_length,
-            self.device,
-        )
-
-        self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
-
-    def batches(self, split="train", nb_to_use=-1, desc=None):
-        assert split in {"train", "test"}
-        input = self.train_input if split == "train" else self.test_input
-        if nb_to_use > 0:
-            input = input[:nb_to_use]
-        if desc is None:
-            desc = f"epoch-{split}"
-        for batch in tqdm.tqdm(
-            input.split(self.batch_size), dynamic_ncols=True, desc=desc
-        ):
-            yield batch
-
-    def vocabulary_size(self):
-        return self.nb_codes
-
-    def produce_results(self, n_epoch, model):
-        with torch.autograd.no_grad():
-            t = model.training
-            model.eval()
-
-            def compute_nb_correct(input, prior_visits):
-                result = input.clone()
-                i = torch.arange(result.size(1), device=result.device)[None, :]
-                ar_mask = (
-                    torch.logical_and(i >= self.prompt_length * 2, i % 2 == 0)
-                    .long()
-                    .expand_as(result)
-                )
-                result *= 1 - ar_mask
-
-                # snake.solver(result,ar_mask)
-
-                masked_inplace_autoregression(
-                    model, self.batch_size, result, ar_mask, device=self.device
-                )
-
-                nb_total = ((prior_visits > 0) * ar_mask).sum()
-
-                nb_correct = (
-                    (result == input).long() * (prior_visits > 0) * ar_mask
-                ).sum()
-
-                # nb_total = result.size(0)
-                # nb_correct = ((result - input).abs().sum(1) == 0).sum()
-
-                return nb_total, nb_correct
-
-            # train_nb_total, train_nb_correct = compute_nb_correct(
-            # self.train_input, self.train_prior_visits
-            # )
-
-            # log_string(
-            # f"accuracy_train nb_total {train_nb_total} nb_correct {train_nb_correct} accuracy {(100.0*train_nb_correct)/train_nb_total:.02f}%"
-            # )
-
-            test_nb_total, test_nb_correct = compute_nb_correct(
-                self.test_input[:1000], self.test_prior_visits[:1000]
-            )
-
-            log_string(
-                f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
-            )
-
-            model.train(t)
-
-
-######################################################################
-
-
-import stack
-
-
-class TaskStack(Task):
-    def __init__(
-        self,
-        nb_train_samples,
-        nb_test_samples,
-        batch_size,
-        nb_steps,
-        nb_stacks,
-        nb_digits,
-        fraction_values_for_train=None,
-        device=torch.device("cpu"),
-    ):
-        self.batch_size = batch_size
-        self.nb_steps = nb_steps
-        self.nb_stacks = nb_stacks
-        self.nb_digits = nb_digits
-        self.device = device
-
-        if fraction_values_for_train is None:
-            values_for_train = None
-            values_for_test = None
-        else:
-            all = torch.randperm(10**nb_digits)
-            nb_for_train = int(all.size(0) * fraction_values_for_train)
-            values_for_train = all[:nb_for_train]
-            values_for_test = all[nb_for_train:]
-
-        self.train_input, self.train_stack_counts = stack.generate_sequences(
-            nb_train_samples,
-            nb_steps,
-            nb_stacks,
-            nb_digits,
-            values_for_train,
-            self.device,
-        )
-
-        self.test_input, self.test_stack_counts = stack.generate_sequences(
-            nb_test_samples,
-            nb_steps,
-            nb_stacks,
-            nb_digits,
-            values_for_test,
-            self.device,
-        )
-
-        i = torch.logical_and(self.test_input % 2 == 1, self.test_input < 2 * nb_stacks)
-        counts = self.test_stack_counts.flatten()[i.flatten()]
-        counts = F.one_hot(counts).sum(0)
-        log_string(f"test_pop_stack_counts {counts}")
-
-        self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
-
-    def batches(self, split="train", nb_to_use=-1, desc=None):
-        assert split in {"train", "test"}
-        input = self.train_input if split == "train" else self.test_input
-        if nb_to_use > 0:
-            input = input[:nb_to_use]
-        if desc is None:
-            desc = f"epoch-{split}"
-        for batch in tqdm.tqdm(
-            input.split(self.batch_size), dynamic_ncols=True, desc=desc
-        ):
-            yield batch
-
-    def vocabulary_size(self):
-        return self.nb_codes
-
-    def produce_results(self, n_epoch, model):
-        with torch.autograd.no_grad():
-            t = model.training
-            model.eval()
-
-            def compute_nb_correct(input):
-                result = input.clone()
-                stack.remove_popped_values(result, self.nb_stacks, self.nb_digits)
-                ar_mask = (result != input).long()
-                masked_inplace_autoregression(
-                    model, self.batch_size, result, ar_mask, device=self.device
-                )
-
-                errors = ((result != input).long() * ar_mask).reshape(
-                    -1, 1 + self.nb_digits
-                )
-                ar_mask = ar_mask.reshape(-1, 1 + self.nb_digits)
-
-                nb_total = ar_mask.max(1).values.sum()
-                nb_correct = nb_total - errors.max(1).values.sum()
-
-                return nb_total, nb_correct
-
-            test_nb_total, test_nb_correct = compute_nb_correct(self.test_input[:1000])
-
-            log_string(
-                f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
-            )
-
-            ##############################################################
-            # Log a few generated sequences
-            input = self.test_input[:10, : 12 * (1 + self.nb_digits)]
-            result = input.clone()
-            stack.remove_popped_values(result, self.nb_stacks, self.nb_digits)
-            ar_mask = (result != input).long()
-            for n in range(result.size(0)):
-                log_string(
-                    f"test_before {stack.seq_to_str(result[n],nb_stacks=self.nb_stacks,nb_digits=self.nb_digits)}"
-                )
-            masked_inplace_autoregression(
-                model, self.batch_size, result, ar_mask, device=self.device
-            )
-            for n in range(result.size(0)):
-                log_string(
-                    f"test_after  {stack.seq_to_str(result[n],nb_stacks=self.nb_stacks,nb_digits=self.nb_digits)}"
-                )
-            ##############################################################
-
-            model.train(t)
-
-
-######################################################################
-
-
-import expr
-
-
-class TaskExpr(Task):
-    def __init__(
-        self,
-        nb_train_samples,
-        nb_test_samples,
-        nb_variables,
-        sequence_length,
-        batch_size,
-        device=torch.device("cpu"),
-    ):
-        self.batch_size = batch_size
-        self.device = device
-
-        train_sequences = expr.generate_sequences(
-            nb_train_samples,
-            nb_variables=nb_variables,
-            length=sequence_length,
-            # length=2 * sequence_length,
-            # randomize_length=True,
-        )
-        test_sequences = expr.generate_sequences(
-            nb_test_samples,
-            nb_variables=nb_variables,
-            length=sequence_length,
-        )
-        self.char2id = dict(
-            [
-                (c, n)
-                for n, c in enumerate(
-                    set("#" + "".join(train_sequences + test_sequences))
-                )
-            ]
-        )
-        self.id2char = dict([(n, c) for c, n in self.char2id.items()])
-
-        self.filler, self.space = self.char2id["#"], self.char2id[" "]
-
-        len_max = max([len(x) for x in train_sequences])
-        self.train_input = torch.cat(
-            [
-                torch.tensor(
-                    [
-                        [self.char2id[c] for c in s + "#" * (len_max - len(s))]
-                        for s in train_sequences
-                    ]
-                )
-            ],
-            0,
-        ).to(device)
-
-        len_max = max([len(x) for x in test_sequences])
-        self.test_input = torch.cat(
-            [
-                torch.tensor(
-                    [
-                        [self.char2id[c] for c in s + "#" * (len_max - len(s))]
-                        for s in test_sequences
-                    ]
-                )
-            ],
-            0,
-        ).to(device)
-
-        self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
-
-    def batches(self, split="train", nb_to_use=-1, desc=None):
-        assert split in {"train", "test"}
-        input = self.train_input if split == "train" else self.test_input
-        if nb_to_use > 0:
-            input = input[:nb_to_use]
-        if desc is None:
-            desc = f"epoch-{split}"
-        for batch in tqdm.tqdm(
-            input.split(self.batch_size), dynamic_ncols=True, desc=desc
-        ):
-            if split == "train":
-                last = (batch != self.filler).max(0).values.nonzero().max() + 1
-                batch = batch[:, :last]
-            yield batch
-
-    def vocabulary_size(self):
-        return self.nb_codes
-
-    def seq2str(self, s):
-        return "".join([self.id2char[k.item()] for k in s])
-
-    def produce_results(self, n_epoch, model):
-        with torch.autograd.no_grad():
-            t = model.training
-            model.eval()
-
-            def compute_nb_correct(input):
-                result = input.clone()
-                ar_mask = (result == self.space).long().cumsum(dim=1).clamp(max=1)
-                result = (1 - ar_mask) * result + ar_mask * self.filler
-                masked_inplace_autoregression(
-                    model, self.batch_size, result, ar_mask, device=self.device
-                )
-
-                nb_total = input.size(0)
-                nb_correct = (input == result).long().min(1).values.sum()
-
-                #######################################################################
-                # Comput predicted vs. true variable values
-
-                nb_delta = torch.zeros(5, dtype=torch.int64)
-                nb_missed = 0
-
-                values_input = expr.extract_results([self.seq2str(s) for s in input])
-                values_result = expr.extract_results([self.seq2str(s) for s in result])
-
-                for i, r in zip(values_input, values_result):
-                    for n, vi in i.items():
-                        vr = r.get(n)
-                        if vr is None or vr < 0:
-                            nb_missed += 1
-                        else:
-                            d = abs(vr - vi)
-                            if d >= nb_delta.size(0):
-                                nb_missed += 1
-                            else:
-                                nb_delta[d] += 1
-
-                ######################################################################
-
-                return nb_total, nb_correct, nb_delta, nb_missed
-
-            (
-                test_nb_total,
-                test_nb_correct,
-                test_nb_delta,
-                test_nb_missed,
-            ) = compute_nb_correct(self.test_input[:1000])
-
-            log_string(
-                f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
-            )
-
-            nb_total = test_nb_delta.sum() + test_nb_missed
-            for d in range(test_nb_delta.size(0)):
-                log_string(
-                    f"error_value {n_epoch} delta {d} {test_nb_delta[d]} {test_nb_delta[d]*100/nb_total:.02f}%"
-                )
-            log_string(
-                f"error_value {n_epoch} missed {test_nb_missed} {test_nb_missed*100/nb_total:.02f}%"
-            )
-
-            ##############################################################
-            # Log a few generated sequences
-            input = self.test_input[:10]
-            result = input.clone()
-            ar_mask = (result == self.space).long().cumsum(dim=1).clamp(max=1)
-            result = (1 - ar_mask) * result + ar_mask * self.filler
-            for n in range(result.size(0)):
-                log_string(f"test_before {self.seq2str(result[n])}")
-            masked_inplace_autoregression(
-                model, self.batch_size, result, ar_mask, device=self.device
-            )
-            correct = (1 - ar_mask) * self.space + ar_mask * input
-            for n in range(result.size(0)):
-                comment = "GOOD" if (result[n] - input[n]).abs().max() == 0 else ""
-                log_string(f"test_after  {self.seq2str(result[n])} {comment}")
-                log_string(f"correct     {self.seq2str(correct[n])}")
-            ##############################################################
-
-            model.train(t)
-
 
 ######################################################################
 
@@ -1204,26 +241,29 @@ picoclvr_pruner_eval = (
 ######################################################################
 
 if args.task == "picoclvr":
-    task = TaskPicoCLVR(
+    task = tasks.PicoCLVR(
         nb_train_samples=args.nb_train_samples,
         nb_test_samples=args.nb_test_samples,
         batch_size=args.batch_size,
         height=args.picoclvr_height,
         width=args.picoclvr_width,
         nb_colors=args.picoclvr_nb_colors,
+        logger=log_string,
         device=device,
         pruner_train=picoclvr_pruner_train,
         pruner_eval=picoclvr_pruner_eval,
     )
 
 elif args.task == "mnist":
-    task = TaskMNIST(
+    task = tasks.MNIST(
+        nb_train_samples=args.nb_train_samples,
+        nb_test_samples=args.nb_test_samples,
         batch_size=args.batch_size,
         device=device,
     )
 
 elif args.task == "maze":
-    task = TaskMaze(
+    task = tasks.Maze(
         nb_train_samples=args.nb_train_samples,
         nb_test_samples=args.nb_test_samples,
         batch_size=args.batch_size,
@@ -1234,7 +274,7 @@ elif args.task == "maze":
     )
 
 elif args.task == "snake":
-    task = TaskSnake(
+    task = tasks.Snake(
         nb_train_samples=args.nb_train_samples,
         nb_test_samples=args.nb_test_samples,
         batch_size=args.batch_size,
@@ -1247,10 +287,11 @@ elif args.task == "snake":
     )
 
 elif args.task == "stack":
-    task = TaskStack(
+    task = tasks.Stack(
         nb_train_samples=args.nb_train_samples,
         nb_test_samples=args.nb_test_samples,
         batch_size=args.batch_size,
+        logger=log_string,
         nb_steps=args.stack_nb_steps,
         nb_stacks=args.stack_nb_stacks,
         nb_digits=args.stack_nb_digits,
@@ -1259,7 +300,7 @@ elif args.task == "stack":
     )
 
 elif args.task == "expr":
-    task = TaskExpr(
+    task = tasks.Expr(
         nb_train_samples=args.nb_train_samples,
         nb_test_samples=args.nb_test_samples,
         nb_variables=args.expr_nb_variables,
@@ -1327,6 +368,8 @@ else:
 
 nb_epochs = args.nb_epochs if args.nb_epochs > 0 else nb_epochs_default
 
+# Compute the entropy of the training tokens
+
 token_count = 0
 for input in task.batches(split="train"):
     token_count += F.one_hot(input, num_classes=task.vocabulary_size()).sum((0, 1))
@@ -1336,6 +379,26 @@ train_set_perplexity = math.exp(entropy)
 
 ##############################
 
+# A bit of paranoia never hurts
+
+train_examples = {}
+
+for input in task.batches(split="train"):
+    assert input.dim()==2 and input.dtype==torch.int64
+    for x in input:
+        train_examples[x.sum().item()]=x
+
+for input in task.batches(split="test"):
+    assert input.dim()==2 and input.dtype==torch.int64
+    for x in input:
+        y = train_examples.get(x.sum().item())
+        if y is not None:
+            assert x.size() != y.size() or (x-y).abs().sum() > 0
+
+del train_examples
+
+##############################
+
 if args.learning_rate_schedule == "cos":
     learning_rate_schedule = {}
     for n_epoch in range(args.nb_epochs):
@@ -1363,7 +426,13 @@ log_string(f"learning_rate_schedule {learning_rate_schedule}")
 nb_samples_seen = 0
 
 if nb_epochs_finished >= nb_epochs:
-    task.produce_results(nb_epochs_finished, model)
+    task.produce_results(
+        nb_epochs_finished,
+        model,
+        args.result_dir,
+        log_string,
+        args.deterministic_synthesis,
+    )
 
 for n_epoch in range(nb_epochs_finished, nb_epochs):
     learning_rate = learning_rate_schedule[n_epoch]
@@ -1415,7 +484,9 @@ for n_epoch in range(nb_epochs_finished, nb_epochs):
             f"perplexity {n_epoch} train_set {train_set_perplexity} train_prediction {train_perplexity} test_prediction {test_perplexity}"
         )
 
-        task.produce_results(n_epoch, model)
+        task.produce_results(
+            n_epoch, model, args.result_dir, log_string, args.deterministic_synthesis
+        )
 
     checkpoint = {
         "nb_epochs_finished": n_epoch + 1,
diff --git a/maze.py b/maze.py
index f6a852e..c2774dd 100755 (executable)
--- a/maze.py
+++ b/maze.py
@@ -13,6 +13,8 @@ v_empty, v_wall, v_start, v_goal, v_path = 0, 1, 2, 3, 4
 
 
 def create_maze(h=11, w=17, nb_walls=8):
+    assert h % 2 == 1 and w % 2 == 1
+
     a, k = 0, 0
 
     while k < nb_walls:
index 6a12a5a..c93010a 100755 (executable)
--- a/mygpt.py
+++ b/mygpt.py
@@ -258,6 +258,30 @@ class MyGPT(nn.Module):
         bs = self.readout(bs)
         return bs
 
+    # ar_mask is a tensor with 0s and 1s, of same shape as input, with
+    # 1s where tokens should be generated. The others are kept
+    # unchanged.
+
+    def masked_inplace_autoregression(
+        self, input, ar_mask, forbidden_tokens=None, deterministic_synthesis=False
+    ):
+        to_generate = (ar_mask.sum(0) > 0).nonzero()
+        if to_generate.min() > 0:
+            self(
+                BracketedSequence(input, 0, to_generate.min())
+            )  # Needed to initialize the model's cache
+        for s in range(to_generate.min(), to_generate.max() + 1):
+            output = self(BracketedSequence(input, s, 1)).x
+            logits = output[:, s]
+            if forbidden_tokens is not None:
+                logits = logits.masked_fill(forbidden_tokens, float("-inf"))
+            if deterministic_synthesis:
+                t_next = logits.argmax(1)
+            else:
+                dist = torch.distributions.categorical.Categorical(logits=logits)
+                t_next = dist.sample()
+            input[:, s] = ar_mask[:, s] * t_next + (1 - ar_mask[:, s]) * input[:, s]
+
 
 ######################################################################
 
@@ -286,8 +310,6 @@ if __name__ == "__main__":
         z = model(BracketedSequence(x, s, 1))
         y2[:, s] = z.x[:, s]
 
-    # print(y1.max(dim = 2).values)
-    # print(y2.max(dim = 2).values)
     print(f"error={((y1 - y2).norm() / (y1.norm() + y2.norm())).item()}")
 
 ######################################################################
diff --git a/tasks.py b/tasks.py
new file mode 100755 (executable)
index 0000000..0f3aaec
--- /dev/null
+++ b/tasks.py
@@ -0,0 +1,1029 @@
+#!/usr/bin/env python
+
+import math, os, tqdm
+
+import torch, torchvision
+
+from torch import nn
+from torch.nn import functional as F
+
+######################################################################
+
+
+def masked_inplace_autoregression(
+    model,
+    batch_size,
+    input,
+    ar_mask,
+    deterministic_synthesis,
+    forbidden_tokens=None,
+    progress_bar_desc="autoregression",
+    device=torch.device("cpu"),
+):
+    batches = zip(input.split(batch_size), ar_mask.split(batch_size))
+
+    if progress_bar_desc is not None:
+        batches = tqdm.tqdm(
+            batches,
+            dynamic_ncols=True,
+            desc=progress_bar_desc,
+            total=input.size(0) // batch_size,
+        )
+
+    for input, ar_mask in batches:
+        model.masked_inplace_autoregression(
+            input, ar_mask, forbidden_tokens, deterministic_synthesis
+        )
+
+
+class Task:
+    def batches(self, split="train"):
+        pass
+
+    def vocabulary_size(self):
+        pass
+
+    def produce_results(
+        self, n_epoch, model, result_dir, logger, deterministic_synthesis
+    ):
+        pass
+
+
+######################################################################
+
+import picoclvr
+
+
+class PicoCLVR(Task):
+    # Make a tensor from a list of strings
+    def tensorize(self, descr):
+        token_descr = [s.strip().split(" ") for s in descr]
+        l = max([len(s) for s in token_descr])
+        token_descr = [s + ["<nul>"] * (l - len(s)) for s in token_descr]
+        id_descr = [[self.token2id[u] for u in s] for s in token_descr]
+        return torch.tensor(id_descr, device=self.device)
+
+    # Make a list of strings from a tensor
+    def detensorize(self, x):
+        return [" ".join([self.id2token[t.item()] for t in r]) for r in x]
+
+    # trim all the tensors in the tuple z to remove as much token from
+    # left and right in the first tensor. If z is a tuple, all its
+    # elements are trimed according to the triming for the first
+    def trim(self, z, token="<nul>"):
+        n = self.token2id[token]
+        if type(z) == tuple:
+            x = z[0]
+            i = (1 - (F.pad(x, (1, 1), value=n) == n).min(0).values.long()).cumsum(0)
+            a, b = (i == 0).nonzero().max(), (i == i.max()).nonzero().min()
+            return tuple([t[:, a:b] for t in z])
+        else:
+            i = (1 - (F.pad(z, (1, 1), value=n) == n).min(0).values.long()).cumsum(0)
+            a, b = (i == 0).nonzero().max(), (i == i.max()).nonzero().min()
+            return z[:, a:b]
+
+    ######################
+    # Not the cleanest part of the code
+
+    # Extract the last image of each sequence, from the last <img>
+    # included, and set to <nul> all the tokens from the beginning of
+    # that image to the end
+    def excise_last_image(self, input):
+        t_img, t_nul = self.token2id["<img>"], self.token2id["<nul>"]
+        nb_img_tokens = self.height * self.width + 1
+
+        input = input.clone()
+        t = (input == t_img).long()
+        tail_masks = (t.cumsum(dim=1) == t.sum(dim=1, keepdim=True)).long()
+        i = (t * tail_masks).nonzero(as_tuple=True)
+        j = (
+            i[0][:, None],
+            i[1][:, None] + torch.arange(nb_img_tokens, device=input.device)[None, :],
+        )
+        images = self.trim(input[j])
+        input[j] = t_nul
+        loss_masks = 1 - tail_masks
+        input, loss_masks = self.trim((input, loss_masks))
+        return input, loss_masks, images
+
+    def add_true_image(self, input, images, loss_masks):
+        t_nul = self.token2id["<nul>"]
+        nb_img_tokens = self.height * self.width + 1
+        input = F.pad(input, (0, nb_img_tokens), value=t_nul)
+        loss_masks = F.pad(loss_masks, (0, nb_img_tokens), value=0)
+        t = (input == t_nul).long()
+        i = (t.cumsum(dim=1) == 1).nonzero(as_tuple=True)
+        j = (
+            i[0][:, None],
+            i[1][:, None] + torch.arange(nb_img_tokens, device=input.device)[None, :],
+        )
+        input[j] = images
+        loss_masks[j] = 1
+        input, loss_masks = self.trim((input, loss_masks))
+        return input, loss_masks
+
+    def add_generated_image(self, input, loss_masks, model, deterministic_synthesis):
+        t_img, t_nul = self.token2id["<img>"], self.token2id["<nul>"]
+        nb_img_tokens = self.height * self.width + 1
+
+        input = F.pad(input, (0, nb_img_tokens), value=t_nul)
+        loss_masks = F.pad(loss_masks, (0, nb_img_tokens), value=0)
+        t = (input == t_nul).long()
+        i = (t.cumsum(dim=1) == 1).nonzero(as_tuple=True)
+        input[i] = t_img
+
+        j = (
+            i[0][:, None],
+            i[1][:, None]
+            + 1
+            + torch.arange(nb_img_tokens - 1, device=input.device)[None, :],
+        )
+        ar_masks = input.new_zeros(input.size(), dtype=torch.int64)
+        ar_masks[j] = 1
+        forbidden_tokens = (
+            torch.arange(self.vocabulary_size(), device=input.device) == t_nul
+        )
+        with torch.autograd.no_grad():
+            t = model.training
+            model.eval()
+            masked_inplace_autoregression(
+                model,
+                self.batch_size,
+                input,
+                ar_masks,
+                deterministic_synthesis,
+                forbidden_tokens,
+                progress_bar_desc=None,
+                device=self.device,
+            )
+            model.train(t)
+
+        input, loss_masks = self.trim((input, loss_masks))
+
+        return input, loss_masks
+
+    ######################
+
+    def __init__(
+        self,
+        nb_train_samples,
+        nb_test_samples,
+        batch_size,
+        height,
+        width,
+        nb_colors=5,
+        logger=None,
+        device=torch.device("cpu"),
+        pruner_train=None,
+        pruner_eval=None,
+    ):
+        def generate_descr(nb, cache_suffix, pruner):
+            return picoclvr.generate(
+                nb,
+                height=self.height,
+                width=self.width,
+                nb_colors=nb_colors,
+                pruner=pruner,
+            )
+
+        self.height = height
+        self.width = width
+        self.batch_size = batch_size
+        self.device = device
+        self.pruner_train = pruner_train
+        self.pruner_eval = pruner_eval
+
+        param = {
+            "nb_train_samples": nb_train_samples,
+            "nb_test_samples": nb_test_samples,
+            "height": height,
+            "width": width,
+            "nb_colors": nb_colors,
+            "batch_size": batch_size,
+            "rng_state": list(torch.get_rng_state()),
+        }
+
+        if logger is not None:
+            logger(
+                f"generating {nb_train_samples+nb_test_samples} samples (can take some time)"
+            )
+
+        self.train_descr = generate_descr(
+            nb_train_samples, "train", pruner=self.pruner_train
+        )
+        self.test_descr = generate_descr(nb_test_samples, "test", pruner=None)
+
+        # Build the tokenizer
+        tokens = {"<nul>", "<img>"}
+        for d in [self.train_descr, self.test_descr]:
+            for s in d:
+                for t in s.strip().split(" "):
+                    tokens.add(t)
+        # make this set a sorted list to get the same tensors given
+        # the same descr
+        tokens = list(tokens)
+        tokens.sort()
+        self.token2id = dict([(t, n) for n, t in enumerate(tokens)])
+        self.id2token = dict([(n, t) for n, t in enumerate(tokens)])
+
+        # Tokenize the train and test sets
+        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"}
+        input = self.train_input if split == "train" else self.test_input
+        for batch in tqdm.tqdm(
+            input.split(self.batch_size), dynamic_ncols=True, desc=f"epoch-{split}"
+        ):
+            yield self.trim(batch)
+
+    def vocabulary_size(self):
+        return len(self.token2id)
+
+    def compute_missing_properties(
+        self, n_epoch, model, logger, deterministic_synthesis, pruner=None
+    ):
+        acc_nb_requested_properties = []
+        acc_nb_missing_properties = []
+        acc_nb_results = 0
+
+        for input in tqdm.tqdm(
+            self.test_input.split(self.batch_size),
+            dynamic_ncols=True,
+            desc=f"test-properties",
+        ):
+            tape, loss_masks, _ = self.excise_last_image(input)
+            tape, loss_masks = self.add_generated_image(
+                tape, loss_masks, model, deterministic_synthesis
+            )
+            result_descr = self.detensorize(tape)
+            np = picoclvr.nb_properties(
+                result_descr,
+                height=self.height,
+                width=self.width,
+                pruner=pruner,
+            )
+            nb_requested_properties, _, nb_missing_properties = zip(*np)
+            acc_nb_requested_properties += nb_requested_properties
+            acc_nb_missing_properties += nb_missing_properties
+            acc_nb_results += len(result_descr)
+
+        nb_requested_properties = sum(acc_nb_requested_properties)
+        nb_missing_properties = sum(acc_nb_missing_properties)
+
+        prefix = "" if pruner is None else "pruned_"
+        logger(f"nb_{prefix}samples {n_epoch} {acc_nb_results}")
+        logger(
+            f"property_{prefix}nb {n_epoch} requested {sum(acc_nb_requested_properties)} missing {sum(acc_nb_missing_properties)}"
+        )
+        logger(
+            f"property_{prefix}miss {n_epoch} {100*nb_missing_properties/nb_requested_properties:.02f}%"
+        )
+
+    ######################################################################
+
+    def produce_results(
+        self, n_epoch, model, result_dir, logger, deterministic_synthesis
+    ):
+        self.compute_missing_properties(n_epoch, model, logger, deterministic_synthesis)
+
+        if self.pruner_eval is not None:
+            self.compute_missing_properties(n_epoch, model, self.pruner_eval)
+
+        nb_tokens_to_generate = self.height * self.width + 3
+        result_descr = []
+        nb_per_primer = 8
+        primer = []
+
+        for primer_descr in [
+            "red above green <sep> green top <sep> blue right of red",
+            "there is red <sep> there is yellow <sep> there is blue",
+            "red below yellow <sep> yellow below green <sep> green below blue <sep> red right <sep> yellow left <sep> green right <sep> blue left",
+            "green bottom <sep> yellow bottom <sep> green left of blue <sep> yellow right of blue <sep> blue top",
+        ]:
+            primer += [primer_descr] * nb_per_primer
+
+        tape = self.tensorize(primer)
+        loss_masks = 1 - (tape == self.token2id["<nul>"]).long()
+        tape, loss_masks = self.add_generated_image(
+            tape, loss_masks, model, deterministic_synthesis
+        )
+        result_descr = self.detensorize(tape)
+
+        np = picoclvr.nb_properties(result_descr, height=self.height, width=self.width)
+
+        acc_nb_requested_properties, _, acc_nb_missing_properties = zip(*np)
+        acc_nb_results = len(result_descr)
+
+        nb_requested_properties = sum(acc_nb_requested_properties)
+        nb_missing_properties = sum(acc_nb_missing_properties)
+
+        prefix = "demo_"
+        logger(f"nb_{prefix}samples {n_epoch} {acc_nb_results}")
+        logger(
+            f"property_{prefix}nb {n_epoch} requested {sum(acc_nb_requested_properties)} missing {sum(acc_nb_missing_properties)}"
+        )
+        logger(
+            f"property_{prefix}miss {n_epoch} {100*nb_missing_properties/nb_requested_properties:.02f}%"
+        )
+
+        img = picoclvr.descr2img(result_descr, height=self.height, width=self.width)
+
+        if img.dim() == 5:
+            if img.size(1) == 1:
+                img = F.pad(img.squeeze(1), pad=(1, 1, 1, 1), value=64)
+            else:
+                img = torch.cat(
+                    [
+                        torchvision.utils.make_grid(x, padding=1, pad_value=64)[None]
+                        for x in img
+                    ],
+                    0,
+                )
+
+        image_name = os.path.join(result_dir, f"picoclvr_result_{n_epoch:04d}.png")
+        torchvision.utils.save_image(
+            img / 255.0, image_name, nrow=nb_per_primer, padding=1, pad_value=0.0
+        )
+        logger(f"wrote {image_name}")
+
+
+######################################################################
+
+
+class MNIST(Task):
+    def __init__(
+        self, nb_train_samples, nb_test_samples, batch_size, device=torch.device("cpu")
+    ):
+        self.nb_train_samples = (nb_train_samples,)
+        self.nb_test_samples = (nb_test_samples,)
+        self.batch_size = batch_size
+        self.device = device
+        data_set = torchvision.datasets.MNIST(root="./data", train=True, download=True)
+        self.train_input = data_set.data[:nb_train_samples].view(-1, 28 * 28).long()
+        data_set = torchvision.datasets.MNIST(root="./data", train=False, download=True)
+        self.test_input = data_set.data[:nb_test_samples].view(-1, 28 * 28).long()
+
+    def batches(self, split="train", nb_to_use=-1, desc=None):
+        assert split in {"train", "test"}
+        input = self.train_input if split == "train" else self.test_input
+        if nb_to_use > 0:
+            input = input[:nb_to_use]
+        if desc is None:
+            desc = f"epoch-{split}"
+        for batch in tqdm.tqdm(
+            input.split(self.batch_size), dynamic_ncols=True, desc=desc
+        ):
+            yield batch
+
+    def vocabulary_size(self):
+        return 256
+
+    def produce_results(
+        self, n_epoch, model, result_dir, logger, deterministic_synthesis
+    ):
+        results = torch.empty(64, 28 * 28, device=self.device, dtype=torch.int64)
+        ar_mask = torch.full_like(results, 1)
+        masked_inplace_autoregression(
+            model,
+            self.batch_size,
+            results,
+            ar_mask,
+            deterministic_synthesis,
+            device=self.device,
+        )
+        image_name = os.path.join(result_dir, f"mnist_result_{n_epoch:04d}.png")
+        torchvision.utils.save_image(
+            1 - results.reshape(-1, 1, 28, 28) / 255.0,
+            image_name,
+            nrow=16,
+            pad_value=0.8,
+        )
+        logger(f"wrote {image_name}")
+
+
+######################################################################
+
+import maze
+
+
+class Maze(Task):
+    def map2seq(self, *m):
+        return torch.cat([x.flatten(1) for x in m], 1)
+
+    def seq2map(self, s):
+        s = s.reshape(s.size(0), -1, self.height, self.width)
+        return (s[:, k] for k in range(s.size(1)))
+
+    def __init__(
+        self,
+        nb_train_samples,
+        nb_test_samples,
+        batch_size,
+        height,
+        width,
+        nb_walls,
+        device=torch.device("cpu"),
+    ):
+        self.batch_size = batch_size
+        self.height = height
+        self.width = width
+        self.device = device
+
+        train_mazes, train_paths, _ = maze.create_maze_data(
+            nb_train_samples,
+            height=height,
+            width=width,
+            nb_walls=nb_walls,
+            progress_bar=lambda x: tqdm.tqdm(x, dynamic_ncols=True, desc=f"data-train"),
+        )
+        self.train_input = self.map2seq(train_mazes.to(device), train_paths.to(device))
+
+        test_mazes, test_paths, _ = maze.create_maze_data(
+            nb_test_samples,
+            height=height,
+            width=width,
+            nb_walls=nb_walls,
+            progress_bar=lambda x: tqdm.tqdm(x, dynamic_ncols=True, desc=f"data-test"),
+        )
+        self.test_input = self.map2seq(test_mazes.to(device), test_paths.to(device))
+
+        self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
+
+    def batches(self, split="train", nb_to_use=-1, desc=None):
+        assert split in {"train", "test"}
+        input = self.train_input if split == "train" else self.test_input
+        if nb_to_use > 0:
+            input = input[:nb_to_use]
+        if desc is None:
+            desc = f"epoch-{split}"
+        for batch in tqdm.tqdm(
+            input.split(self.batch_size), dynamic_ncols=True, desc=desc
+        ):
+            yield batch
+
+    def vocabulary_size(self):
+        return self.nb_codes
+
+    def compute_error(
+        self, model, split="train", nb_to_use=-1, deterministic_synthesis=False
+    ):
+        nb_total, nb_correct = 0, 0
+        count = torch.zeros(
+            self.width * self.height,
+            self.width * self.height,
+            device=self.device,
+            dtype=torch.int64,
+        )
+
+        for input in self.batches(split, nb_to_use):
+            result = input.clone()
+            ar_mask = result.new_zeros(result.size())
+            ar_mask[:, self.height * self.width :] = 1
+            result *= 1 - ar_mask
+            masked_inplace_autoregression(
+                model,
+                self.batch_size,
+                result,
+                ar_mask,
+                deterministic_synthesis,
+                progress_bar_desc=None,
+                device=self.device,
+            )
+            mazes, paths = self.seq2map(result)
+            path_correctness = maze.path_correctness(mazes, paths)
+            nb_correct += path_correctness.long().sum()
+            nb_total += mazes.size(0)
+
+            optimal_path_lengths = (
+                (input[:, self.height * self.width :] == maze.v_path).long().sum(1)
+            )
+            predicted_path_lengths = (
+                (result[:, self.height * self.width :] == maze.v_path).long().sum(1)
+            )
+            optimal_path_lengths = optimal_path_lengths[path_correctness]
+            predicted_path_lengths = predicted_path_lengths[path_correctness]
+            count[optimal_path_lengths, predicted_path_lengths] += 1
+
+        if count.max() == 0:
+            count = None
+        else:
+            count = count[
+                : count.sum(1).nonzero().max() + 1, : count.sum(0).nonzero().max() + 1
+            ]
+
+        return nb_total, nb_correct, count
+
+    def produce_results(
+        self, n_epoch, model, result_dir, logger, deterministic_synthesis
+    ):
+        with torch.autograd.no_grad():
+            t = model.training
+            model.eval()
+
+            train_nb_total, train_nb_correct, count = self.compute_error(
+                model,
+                "train",
+                nb_to_use=1000,
+                deterministic_synthesis=deterministic_synthesis,
+            )
+            logger(
+                f"accuracy_train {n_epoch} nb_total {train_nb_total} nb_correct {train_nb_correct} accuracy {(100.0*train_nb_correct)/train_nb_total:.02f}%"
+            )
+
+            test_nb_total, test_nb_correct, count = self.compute_error(
+                model,
+                "test",
+                nb_to_use=1000,
+                deterministic_synthesis=deterministic_synthesis,
+            )
+            logger(
+                f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
+            )
+
+            if count is not None:
+                proportion_optimal = count.diagonal().sum().float() / count.sum()
+                logger(f"proportion_optimal_test {proportion_optimal*100:.02f}%")
+                with open(
+                    os.path.join(result_dir, f"maze_result_{n_epoch:04d}.txt"), "w"
+                ) as f:
+                    for i in range(count.size(0)):
+                        for j in range(count.size(1)):
+                            eol = " " if j < count.size(1) - 1 else "\n"
+                            f.write(f"{count[i,j]}{eol}")
+
+            input = self.test_input[:48]
+            result = input.clone()
+            ar_mask = result.new_zeros(result.size())
+            ar_mask[:, self.height * self.width :] = 1
+            result *= 1 - ar_mask
+            masked_inplace_autoregression(
+                model,
+                self.batch_size,
+                result,
+                ar_mask,
+                deterministic_synthesis,
+                device=self.device,
+            )
+
+            mazes, paths = self.seq2map(input)
+            _, predicted_paths = self.seq2map(result)
+
+            filename = os.path.join(result_dir, f"maze_result_{n_epoch:04d}.png")
+            maze.save_image(
+                filename,
+                mazes=mazes,
+                target_paths=paths,
+                predicted_paths=predicted_paths,
+                path_correct=maze.path_correctness(mazes, predicted_paths),
+                path_optimal=maze.path_optimality(paths, predicted_paths),
+            )
+            logger(f"wrote {filename}")
+
+            model.train(t)
+
+
+######################################################################
+
+
+import snake
+
+
+class Snake(Task):
+    def __init__(
+        self,
+        nb_train_samples,
+        nb_test_samples,
+        batch_size,
+        height,
+        width,
+        nb_colors,
+        length,
+        prompt_length,
+        device=torch.device("cpu"),
+    ):
+        self.batch_size = batch_size
+        self.height = height
+        self.width = width
+        self.device = device
+        self.prompt_length = prompt_length
+
+        self.train_input, self.train_prior_visits, _, _ = snake.generate_sequences(
+            nb_train_samples,
+            height,
+            width,
+            nb_colors,
+            length,
+            prompt_length,
+            self.device,
+        )
+        self.test_input, self.test_prior_visits, _, _ = snake.generate_sequences(
+            nb_test_samples,
+            height,
+            width,
+            nb_colors,
+            length,
+            prompt_length,
+            self.device,
+        )
+
+        self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
+
+    def batches(self, split="train", nb_to_use=-1, desc=None):
+        assert split in {"train", "test"}
+        input = self.train_input if split == "train" else self.test_input
+        if nb_to_use > 0:
+            input = input[:nb_to_use]
+        if desc is None:
+            desc = f"epoch-{split}"
+        for batch in tqdm.tqdm(
+            input.split(self.batch_size), dynamic_ncols=True, desc=desc
+        ):
+            yield batch
+
+    def vocabulary_size(self):
+        return self.nb_codes
+
+    def produce_results(
+        self, n_epoch, model, result_dir, logger, deterministic_synthesis
+    ):
+        with torch.autograd.no_grad():
+            t = model.training
+            model.eval()
+
+            def compute_nb_correct(input, prior_visits):
+                result = input.clone()
+                i = torch.arange(result.size(1), device=result.device)[None, :]
+                ar_mask = (
+                    torch.logical_and(i >= self.prompt_length * 2, i % 2 == 0)
+                    .long()
+                    .expand_as(result)
+                )
+                result *= 1 - ar_mask
+
+                # snake.solver(result,ar_mask)
+
+                masked_inplace_autoregression(
+                    model,
+                    self.batch_size,
+                    result,
+                    ar_mask,
+                    deterministic_synthesis,
+                    device=self.device,
+                )
+
+                nb_total = ((prior_visits > 0) * ar_mask).sum()
+
+                nb_correct = (
+                    (result == input).long() * (prior_visits > 0) * ar_mask
+                ).sum()
+
+                # nb_total = result.size(0)
+                # nb_correct = ((result - input).abs().sum(1) == 0).sum()
+
+                return nb_total, nb_correct
+
+            # train_nb_total, train_nb_correct = compute_nb_correct(
+            # self.train_input, self.train_prior_visits
+            # )
+
+            # logger(
+            # f"accuracy_train nb_total {train_nb_total} nb_correct {train_nb_correct} accuracy {(100.0*train_nb_correct)/train_nb_total:.02f}%"
+            # )
+
+            test_nb_total, test_nb_correct = compute_nb_correct(
+                self.test_input[:1000], self.test_prior_visits[:1000]
+            )
+
+            logger(
+                f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
+            )
+
+            model.train(t)
+
+
+######################################################################
+
+
+import stack
+
+
+class Stack(Task):
+    def __init__(
+        self,
+        nb_train_samples,
+        nb_test_samples,
+        batch_size,
+        logger,
+        nb_steps,
+        nb_stacks,
+        nb_digits,
+        fraction_values_for_train=None,
+        device=torch.device("cpu"),
+    ):
+        self.batch_size = batch_size
+        self.nb_steps = nb_steps
+        self.nb_stacks = nb_stacks
+        self.nb_digits = nb_digits
+        self.device = device
+
+        if fraction_values_for_train is None:
+            values_for_train = None
+            values_for_test = None
+        else:
+            all = torch.randperm(10**nb_digits)
+            nb_for_train = int(all.size(0) * fraction_values_for_train)
+            values_for_train = all[:nb_for_train]
+            values_for_test = all[nb_for_train:]
+
+        self.train_input, self.train_stack_counts = stack.generate_sequences(
+            nb_train_samples,
+            nb_steps,
+            nb_stacks,
+            nb_digits,
+            values_for_train,
+            self.device,
+        )
+
+        self.test_input, self.test_stack_counts = stack.generate_sequences(
+            nb_test_samples,
+            nb_steps,
+            nb_stacks,
+            nb_digits,
+            values_for_test,
+            self.device,
+        )
+
+        i = torch.logical_and(self.test_input % 2 == 1, self.test_input < 2 * nb_stacks)
+        counts = self.test_stack_counts.flatten()[i.flatten()]
+        counts = F.one_hot(counts).sum(0)
+        logger(f"test_pop_stack_counts {counts}")
+
+        self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
+
+    def batches(self, split="train", nb_to_use=-1, desc=None):
+        assert split in {"train", "test"}
+        input = self.train_input if split == "train" else self.test_input
+        if nb_to_use > 0:
+            input = input[:nb_to_use]
+        if desc is None:
+            desc = f"epoch-{split}"
+        for batch in tqdm.tqdm(
+            input.split(self.batch_size), dynamic_ncols=True, desc=desc
+        ):
+            yield batch
+
+    def vocabulary_size(self):
+        return self.nb_codes
+
+    def produce_results(
+        self, n_epoch, model, result_dir, logger, deterministic_synthesis
+    ):
+        with torch.autograd.no_grad():
+            t = model.training
+            model.eval()
+
+            def compute_nb_correct(input):
+                result = input.clone()
+                stack.remove_popped_values(result, self.nb_stacks, self.nb_digits)
+                ar_mask = (result != input).long()
+                masked_inplace_autoregression(
+                    model,
+                    self.batch_size,
+                    result,
+                    ar_mask,
+                    deterministic_synthesis,
+                    device=self.device,
+                )
+
+                errors = ((result != input).long() * ar_mask).reshape(
+                    -1, 1 + self.nb_digits
+                )
+                ar_mask = ar_mask.reshape(-1, 1 + self.nb_digits)
+
+                nb_total = ar_mask.max(1).values.sum()
+                nb_correct = nb_total - errors.max(1).values.sum()
+
+                return nb_total, nb_correct
+
+            test_nb_total, test_nb_correct = compute_nb_correct(self.test_input[:1000])
+
+            logger(
+                f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
+            )
+
+            ##############################################################
+            # Log a few generated sequences
+            input = self.test_input[:10, : 12 * (1 + self.nb_digits)]
+            result = input.clone()
+            stack.remove_popped_values(result, self.nb_stacks, self.nb_digits)
+            ar_mask = (result != input).long()
+            for n in range(result.size(0)):
+                logger(
+                    f"test_before {stack.seq_to_str(result[n],nb_stacks=self.nb_stacks,nb_digits=self.nb_digits)}"
+                )
+                masked_inplace_autoregression(
+                    model,
+                    self.batch_size,
+                    result,
+                    ar_mask,
+                    deterministic_synthesis,
+                    device=self.device,
+                )
+            for n in range(result.size(0)):
+                logger(
+                    f"test_after  {stack.seq_to_str(result[n],nb_stacks=self.nb_stacks,nb_digits=self.nb_digits)}"
+                )
+            ##############################################################
+
+            model.train(t)
+
+
+######################################################################
+
+
+import expr
+
+
+class Expr(Task):
+    def __init__(
+        self,
+        nb_train_samples,
+        nb_test_samples,
+        nb_variables,
+        sequence_length,
+        batch_size,
+        device=torch.device("cpu"),
+    ):
+        self.batch_size = batch_size
+        self.device = device
+
+        train_sequences = expr.generate_sequences(
+            nb_train_samples,
+            nb_variables=nb_variables,
+            length=sequence_length,
+            # length=2 * sequence_length,
+            # randomize_length=True,
+        )
+        test_sequences = expr.generate_sequences(
+            nb_test_samples,
+            nb_variables=nb_variables,
+            length=sequence_length,
+        )
+        self.char2id = dict(
+            [
+                (c, n)
+                for n, c in enumerate(
+                    set("#" + "".join(train_sequences + test_sequences))
+                )
+            ]
+        )
+        self.id2char = dict([(n, c) for c, n in self.char2id.items()])
+
+        self.filler, self.space = self.char2id["#"], self.char2id[" "]
+
+        len_max = max([len(x) for x in train_sequences])
+        self.train_input = torch.cat(
+            [
+                torch.tensor(
+                    [
+                        [self.char2id[c] for c in s + "#" * (len_max - len(s))]
+                        for s in train_sequences
+                    ]
+                )
+            ],
+            0,
+        ).to(device)
+
+        len_max = max([len(x) for x in test_sequences])
+        self.test_input = torch.cat(
+            [
+                torch.tensor(
+                    [
+                        [self.char2id[c] for c in s + "#" * (len_max - len(s))]
+                        for s in test_sequences
+                    ]
+                )
+            ],
+            0,
+        ).to(device)
+
+        self.nb_codes = max(self.train_input.max(), self.test_input.max()) + 1
+
+    def batches(self, split="train", nb_to_use=-1, desc=None):
+        assert split in {"train", "test"}
+        input = self.train_input if split == "train" else self.test_input
+        if nb_to_use > 0:
+            input = input[:nb_to_use]
+        if desc is None:
+            desc = f"epoch-{split}"
+        for batch in tqdm.tqdm(
+            input.split(self.batch_size), dynamic_ncols=True, desc=desc
+        ):
+            if split == "train":
+                last = (batch != self.filler).max(0).values.nonzero().max() + 3
+                batch = batch[:, :last]
+            yield batch
+
+    def vocabulary_size(self):
+        return self.nb_codes
+
+    def seq2str(self, s):
+        return "".join([self.id2char[k.item()] for k in s])
+
+    def produce_results(
+        self, n_epoch, model, result_dir, logger, deterministic_synthesis
+    ):
+        with torch.autograd.no_grad():
+            t = model.training
+            model.eval()
+
+            def compute_nb_correct(input):
+                result = input.clone()
+                ar_mask = (result == self.space).long().cumsum(dim=1).clamp(max=1)
+                result = (1 - ar_mask) * result + ar_mask * self.filler
+                masked_inplace_autoregression(
+                    model,
+                    self.batch_size,
+                    result,
+                    ar_mask,
+                    deterministic_synthesis,
+                    device=self.device,
+                )
+
+                nb_total = input.size(0)
+                nb_correct = (input == result).long().min(1).values.sum()
+
+                #######################################################################
+                # Comput predicted vs. true variable values
+
+                nb_delta = torch.zeros(5, dtype=torch.int64)
+                nb_missed = 0
+
+                values_input = expr.extract_results([self.seq2str(s) for s in input])
+                values_result = expr.extract_results([self.seq2str(s) for s in result])
+
+                for i, r in zip(values_input, values_result):
+                    for n, vi in i.items():
+                        vr = r.get(n)
+                        if vr is None or vr < 0:
+                            nb_missed += 1
+                        else:
+                            d = abs(vr - vi)
+                            if d >= nb_delta.size(0):
+                                nb_missed += 1
+                            else:
+                                nb_delta[d] += 1
+
+                ######################################################################
+
+                return nb_total, nb_correct, nb_delta, nb_missed
+
+            (
+                test_nb_total,
+                test_nb_correct,
+                test_nb_delta,
+                test_nb_missed,
+            ) = compute_nb_correct(self.test_input[:1000])
+
+            logger(
+                f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
+            )
+
+            nb_total = test_nb_delta.sum() + test_nb_missed
+            for d in range(test_nb_delta.size(0)):
+                logger(
+                    f"error_value {n_epoch} delta {d} {test_nb_delta[d]} {test_nb_delta[d]*100/nb_total:.02f}%"
+                )
+            logger(
+                f"error_value {n_epoch} missed {test_nb_missed} {test_nb_missed*100/nb_total:.02f}%"
+            )
+
+            ##############################################################
+            # Log a few generated sequences
+            input = self.test_input[:10]
+            result = input.clone()
+            ar_mask = (result == self.space).long().cumsum(dim=1).clamp(max=1)
+            result = (1 - ar_mask) * result + ar_mask * self.filler
+            for n in range(result.size(0)):
+                logger(f"test_before {self.seq2str(result[n])}")
+                masked_inplace_autoregression(
+                    model,
+                    self.batch_size,
+                    result,
+                    ar_mask,
+                    deterministic_synthesis,
+                    device=self.device,
+                )
+            correct = (1 - ar_mask) * self.space + ar_mask * input
+            for n in range(result.size(0)):
+                comment = "GOOD" if (result[n] - input[n]).abs().max() == 0 else ""
+                logger(f"test_after  {self.seq2str(result[n])} {comment}")
+                logger(f"correct     {self.seq2str(correct[n])}")
+            ##############################################################
+
+            model.train(t)
+
+
+######################################################################