vegans.utils.loading package

Subpackages

• vegans.utils.loading.architectures package
  • Submodules
  • vegans.utils.loading.architectures.celeba module
  • vegans.utils.loading.architectures.example module
  • vegans.utils.loading.architectures.mnist module
  • Module contents

Submodules

vegans.utils.loading.CIFAR100Loader module

class vegans.utils.loading.CIFAR100Loader.CIFAR100Loader(root=None)

Bases: vegans.utils.loading.CIFAR10Loader.CIFAR10Loader

static _preprocess(X_train, y_train, X_test, y_test)

Preprocess mnist by normalizing and padding.

vegans.utils.loading.CIFAR10Loader module

class vegans.utils.loading.CIFAR10Loader.CIFAR10Loader(root=None)

Bases: vegans.utils.loading.MNISTLoader.MNISTLoader

static _preprocess(X_train, y_train, X_test, y_test)

Preprocess mnist by normalizing and padding.

load_adversary(x_dim=(3, 32, 32), y_dim=10, adv_type='Discriminator')

Loads a working adversary architecture

load_autoencoder(z_dim=64, y_dim=10)

Loads a working autoencoder architecture

load_decoder(z_dim=64, y_dim=10)

Loads a working generator architecture

load_encoder(x_dim=(3, 32, 32), z_dim=64, y_dim=10)

Loads a working encoder architecture

load_generator(x_dim=(3, 32, 32), z_dim=64, y_dim=10)

Loads a working generator architecture

vegans.utils.loading.CelebALoader module

class vegans.utils.loading.CelebALoader.CelebALoader(root=None, batch_size=32, max_loaded_images=5000, crop_size=128, output_shape=64, verbose=False, **kwargs)

Bases: vegans.utils.loading.DatasetLoader.DatasetLoader

__init__(root=None, batch_size=32, max_loaded_images=5000, crop_size=128, output_shape=64, verbose=False, **kwargs)

Parameters

• batch_size (int) – batch size during training.

• max_loaded_images (int) – Number of examples loaded into memory, before new batch is loaded.

• kwargs – Other input arguments to torchvision.utils.data.DataLoader

load_adversary(x_dim=None, y_dim=40, adv_type='Discriminator')

Loads a working adversary architecture

load_autoencoder(x_dim=None, y_dim=40)

Loads a working autoencoder architecture

load_decoder(x_dim=None, z_dim=(16, 4, 4), y_dim=40)

Loads a working generator architecture

load_encoder(x_dim=None, z_dim=(16, 4, 4), y_dim=40)

Loads a working encoder architecture

load_generator(x_dim=None, z_dim=(16, 4, 4), y_dim=40)

Loads a working generator architecture

vegans.utils.loading.DatasetLoader module

class vegans.utils.loading.DatasetLoader.DatasetLoader(metadata, root=None)

Bases: abc.ABC

Class that downloads a dataset and caches it locally. Assumes that the file can be downloaded (i.e. publicly available via an URI)

So far available are:

• MNIST: Handwritten digits with labels. Can be downloaded via download=True.

• FashionMNIST: Clothes with labels. Can be downloaded via download=True.

• CelebA: Pictures of celebrities with attributes. Must be downloaded from https://www.kaggle.com/jessicali9530/celeba-dataset

  and moved into root folder.

• CIFAR: Pictures of objects with labels. Must be downloaded from http://www.cs.toronto.edu/~kriz/cifar.html

  and moved into root folder.

_check_dataset_integrity_or_raise(path, expected_hash)

Ensures that the dataset exists and its MD5 checksum matches the expected hash.

abstract _load_from_disk()

Given a Path to the file and a DataLoaderMetadata object, returns train and test sets as numpy arrays. One can assume that the file exists and its MD5 checksum has been verified before this function is called

load()

Load the dataset in memory, as numpy arrays. Downloads the dataset if it is not present _is_already_downloaded

abstract load_adversary()

Loads a working adversary architecture

abstract load_autoencoder()

Loads a working autoencoder architecture

abstract load_decoder()

Loads a working generator architecture

abstract load_encoder()

Loads a working encoder architecture

abstract load_generator()

Loads a working generator architecture

class vegans.utils.loading.DatasetLoader.DatasetMetaData(directory, m5hashes)

Bases: object

vegans.utils.loading.ExampleLoader module

class vegans.utils.loading.ExampleLoader.ExampleLoader

Bases: vegans.utils.loading.DatasetLoader.DatasetLoader

load_adversary(x_dim, y_dim=None, adv_type='Discriminator')

Loads a working adversary architecture

load_autoencoder(x_dim, y_dim=None)

Loads a working autoencoder architecture

load_decoder(x_dim, z_dim, y_dim=None)

Loads a working generator architecture

load_encoder(x_dim, z_dim, y_dim=None)

Loads a working encoder architecture

load_generator(x_dim, z_dim, y_dim=None)

Loads a working generator architecture

vegans.utils.loading.FashionMNISTLoader module

class vegans.utils.loading.FashionMNISTLoader.FashionMNISTLoader(root=None)

Bases: vegans.utils.loading.MNISTLoader.MNISTLoader

vegans.utils.loading.MNISTLoader module

class vegans.utils.loading.MNISTLoader.MNISTLoader(root=None)

Bases: vegans.utils.loading.DatasetLoader.DatasetLoader

static _preprocess(X_train, y_train, X_test, y_test)

Preprocess mnist by normalizing and padding.

load_adversary(x_dim=(1, 32, 32), y_dim=10, adv_type='Discriminator')

Loads a working adversary architecture

load_autoencoder(x_dim=(1, 32, 32), y_dim=10)

Loads a working autoencoder architecture

load_decoder(x_dim=(1, 32, 32), z_dim=32, y_dim=10)

Loads a working generator architecture

load_encoder(x_dim=(1, 32, 32), z_dim=32, y_dim=10)

Loads a working encoder architecture

load_generator(x_dim=(1, 32, 32), z_dim=32, y_dim=10)

Loads a working generator architecture

Module contents