Module `pysimt.datasets.base`

Expand source code

import json
from abc import ABCMeta, abstractmethod


class BaseDataset(metaclass=ABCMeta):
    r"""Abstract dataset class that all nmtpytorch datasets should be
    extending.

    A sane enough `__repr__` is provided which will neatly dump the
    instance attributes not prefixed with `_`. So make sure to hide
    non-public attributes by prefixing them with `_` to avoid cluttering
    on the terminal. You can also redefine this method in the deriving
    classes.

    """
    requires_vocab = False

    @abstractmethod
    def __len__(self):
        pass

    @abstractmethod
    def __getitem__(self, idx):
        """Returns the `idx`'th item of the dataset."""
        pass

    @abstractmethod
    def collate(self, elems):
        """Collates a list of elements into a `torch.Tensor`."""
        pass

    def get_batch_tensor(self, idxs):
        """Collates by looking up elements of the `idxs` list."""
        return self.collate([self.__getitem__(idx) for idx in idxs])

    def __repr__(self):
        public_attrs = [s for s in self.__dict__.keys() if s[0] != '_']
        info_dict = {k: self.__dict__[k] for k in public_attrs}
        public_str = json.dumps(
            info_dict, indent=1, default=lambda obj: str(obj))
        return f"\n{self.__class__.__name__} {public_str}\n"

Classes

class BaseDataset

Abstract dataset class that all nmtpytorch datasets should be extending.

A sane enough __repr__ is provided which will neatly dump the instance attributes not prefixed with _. So make sure to hide non-public attributes by prefixing them with _ to avoid cluttering on the terminal. You can also redefine this method in the deriving classes.

Expand source code

class BaseDataset(metaclass=ABCMeta):
    r"""Abstract dataset class that all nmtpytorch datasets should be
    extending.

    A sane enough `__repr__` is provided which will neatly dump the
    instance attributes not prefixed with `_`. So make sure to hide
    non-public attributes by prefixing them with `_` to avoid cluttering
    on the terminal. You can also redefine this method in the deriving
    classes.

    """
    requires_vocab = False

    @abstractmethod
    def __len__(self):
        pass

    @abstractmethod
    def __getitem__(self, idx):
        """Returns the `idx`'th item of the dataset."""
        pass

    @abstractmethod
    def collate(self, elems):
        """Collates a list of elements into a `torch.Tensor`."""
        pass

    def get_batch_tensor(self, idxs):
        """Collates by looking up elements of the `idxs` list."""
        return self.collate([self.__getitem__(idx) for idx in idxs])

    def __repr__(self):
        public_attrs = [s for s in self.__dict__.keys() if s[0] != '_']
        info_dict = {k: self.__dict__[k] for k in public_attrs}
        public_str = json.dumps(
            info_dict, indent=1, default=lambda obj: str(obj))
        return f"\n{self.__class__.__name__} {public_str}\n"

Subclasses

ImageFolderDataset

Class variables

var requires_vocab

Methods

def collate(self, elems)

Collates a list of elements into a torch.Tensor.

Expand source code

@abstractmethod
def collate(self, elems):
    """Collates a list of elements into a `torch.Tensor`."""
    pass

def get_batch_tensor(self, idxs)

Collates by looking up elements of the idxs list.

Expand source code

def get_batch_tensor(self, idxs):
    """Collates by looking up elements of the `idxs` list."""
    return self.collate([self.__getitem__(idx) for idx in idxs])