salad.datasets.transforms package

salad.datasets.transforms.tensor(data, dtype=None, device=None, requires_grad=False) → Tensor

Constructs a tensor with data.

Warning

torch.tensor() always copies data. If you have a Tensor data and want to avoid a copy, use torch.Tensor.requires_grad_() or torch.Tensor.detach(). If you have a NumPy ndarray and want to avoid a copy, use torch.from_numpy().

Parameters:

• data (array_like) – Initial data for the tensor. Can be a list, tuple, NumPy ndarray, scalar, and other types.
• dtype (torch.dtype, optional) – the desired data type of returned tensor. Default: if None, infers data type from data.
• device (torch.device, optional) – the desired device of returned tensor. Default: if None, uses the current device for the default tensor type (see torch.set_default_tensor_type()). device will be the CPU for CPU tensor types and the current CUDA device for CUDA tensor types.
• requires_grad (bool, optional) – If autograd should record operations on the returned tensor. Default: False.

Example:

>>> torch.tensor([[0.1, 1.2], [2.2, 3.1], [4.9, 5.2]])
tensor([[ 0.1000,  1.2000],
        [ 2.2000,  3.1000],
        [ 4.9000,  5.2000]])

>>> torch.tensor([0, 1])  # Type inference on data
tensor([ 0,  1])

>>> torch.tensor([[0.11111, 0.222222, 0.3333333]],
                 dtype=torch.float64,
                 device=torch.device('cuda:0'))  # creates a torch.cuda.DoubleTensor
tensor([[ 0.1111,  0.2222,  0.3333]], dtype=torch.float64, device='cuda:0')

>>> torch.tensor(3.14159)  # Create a scalar (zero-dimensional tensor)
tensor(3.1416)

>>> torch.tensor([])  # Create an empty tensor (of size (0,))
tensor([])

Submodules

salad.datasets.transforms.digits module

Standard Transformations for Digit datasets

salad.datasets.transforms.digits.default_normalization(key)

salad.datasets.transforms.digits.default_transforms(key)

salad.datasets.transforms.digits.tensor(data, dtype=None, device=None, requires_grad=False) → Tensor

Constructs a tensor with data.

Warning

torch.tensor() always copies data. If you have a Tensor data and want to avoid a copy, use torch.Tensor.requires_grad_() or torch.Tensor.detach(). If you have a NumPy ndarray and want to avoid a copy, use torch.from_numpy().

Parameters:

• data (array_like) – Initial data for the tensor. Can be a list, tuple, NumPy ndarray, scalar, and other types.
• dtype (torch.dtype, optional) – the desired data type of returned tensor. Default: if None, infers data type from data.
• device (torch.device, optional) – the desired device of returned tensor. Default: if None, uses the current device for the default tensor type (see torch.set_default_tensor_type()). device will be the CPU for CPU tensor types and the current CUDA device for CUDA tensor types.
• requires_grad (bool, optional) – If autograd should record operations on the returned tensor. Default: False.

Example:

>>> torch.tensor([[0.1, 1.2], [2.2, 3.1], [4.9, 5.2]])
tensor([[ 0.1000,  1.2000],
        [ 2.2000,  3.1000],
        [ 4.9000,  5.2000]])

>>> torch.tensor([0, 1])  # Type inference on data
tensor([ 0,  1])

>>> torch.tensor([[0.11111, 0.222222, 0.3333333]],
                 dtype=torch.float64,
                 device=torch.device('cuda:0'))  # creates a torch.cuda.DoubleTensor
tensor([[ 0.1111,  0.2222,  0.3333]], dtype=torch.float64, device='cuda:0')

>>> torch.tensor(3.14159)  # Create a scalar (zero-dimensional tensor)
tensor(3.1416)

>>> torch.tensor([])  # Create an empty tensor (of size (0,))
tensor([])

salad.datasets.transforms.ensembling module

class salad.datasets.transforms.ensembling.Augmentation(dataset, n_samples=1)

Bases: object

class salad.datasets.transforms.ensembling.ImageAugmentation(hflip, xlat_range, affine_std, rot_std=0.0, intens_flip=False, intens_scale_range_lower=None, intens_scale_range_upper=None, intens_offset_range_lower=None, intens_offset_range_upper=None, scale_x_range=None, scale_y_range=None, scale_u_range=None, gaussian_noise_std=0.0, blur_range=None)

Bases: object

augment(X)

augment_pair(X)

salad.datasets.transforms.ensembling.cat_nx2x3(a, b)

Multiply the N 2x3 transformations stored in a with those in b :param a: transformation matrices, (N,2,3) array :param b: transformation matrices, (N,2,3) array :return: a . b

salad.datasets.transforms.ensembling.centre_xf(xf, size)

Centre the transformations in xf around (0,0), where the current centre is assumed to be at the centre of an image of shape size :param xf: transformation matrices, (N,2,3) array :param size: image size :return: centred transformation matrices, (N,2,3) array

salad.datasets.transforms.ensembling.identity_xf(N)

Construct N identity 2x3 transformation matrices :return: array of shape (N, 2, 3)

salad.datasets.transforms.ensembling.inv_nx2x2(X)

Invert the N 2x2 transformation matrices stored in X; a (N,2,2) array :param X: transformation matrices to invert, (N,2,2) array :return: inverse of X

salad.datasets.transforms.ensembling.inv_nx2x3(m)

Invert the N 2x3 transformation matrices stored in X; a (N,2,3) array :param X: transformation matrices to invert, (N,2,3) array :return: inverse of X

salad.datasets.transforms.ensembling.rotation_matrices(thetas)

Generate rotation matrices :param thetas: rotation angles in radians as a (N,) array :return: rotation matrices, (N,2,3) array

salad.datasets.transforms.noise module

class salad.datasets.transforms.noise.DomainConfusion(transform_list, intermediate)

Bases: object

Given x and a set of possible transforms, applies a random transform on x and returns a pair (x, d)

class salad.datasets.transforms.noise.DomainLabel(domain, n_domains)

Bases: object

concats a domain label to the dataset

class salad.datasets.transforms.noise.Gaussian(mu=0.0, sigma=0.1)

Bases: object

Add gaussian noise

class salad.datasets.transforms.noise.InvertContrast

Bases: object

class salad.datasets.transforms.noise.SaltAndPepper(p=0.05)

Bases: object

Adds salt and pepper noise with probability p to a given image or batch of images

class salad.datasets.transforms.noise.Shift(w=5, h=5)

Bases: object

class salad.datasets.transforms.noise.Uniform(p=0.05)

Bases: object

Add uniform noise

salad.datasets.transforms.noisy module

salad.datasets.transforms.noisy.clean_transform(img_size=64)

salad.datasets.transforms.noisy.load_noisy_mnist(path, train=True)

salad.datasets.transforms.noisy.noisy_transform(img_size=64)

salad.datasets.transforms.noisy.transform(x)