tltorch.FactorizedTensor¶

class tltorch.FactorizedTensor(*args, **kwargs)[source]¶

Tensor in Factorized form

Important

All tensor factorization must have an order parameter

Attributes:

decomposition

Returns the factors and parameters composing the tensor in factorized form

name

Factorization name (‘tucker’, ‘tt’, ‘cp’, …)

ndim

Order of the tensor

tensor_shape

Methods

dim()	Order of the tensor
forward([indices])	To use a tensor factorization within a network, use tensor.forward, or, equivalently, tensor()
from_tensor(tensor, rank[, factorization])	Create a factorized tensor by decomposing a dense tensor
new(shape[, rank, factorization])	Main way to create a factorized tensor
normal_([mean, std])	Inialize the factors of the factorization such that the reconstruction follows a Gaussian distribution
size([index])	shape of the tensor
to_tensor()	Reconstruct the full tensor from its factorized form

numel

classmethod new(shape, rank='same', factorization='Tucker', **kwargs)[source]¶

Main way to create a factorized tensor

Parameters:

shapetuple[int]

shape of the factorized tensor to create

rankint, ‘same’ or float, default is ‘same’

rank of the decomposition

factorization{‘CP’, ‘TT’, ‘Tucker’}, optional

Tensor factorization to use to decompose the tensor, by default ‘Tucker’

Returns:

TensorFactorization

Tensor in Factorized form.

Raises:

ValueError

If the factorization given does not exist.

Examples

Create a Tucker tensor of shape (3, 4, 2) with half the parameters as a dense tensor would:

>>> tucker_tensor = FactorizedTensor.new((3, 4, 2)), rank=0.5, factorization='tucker')

classmethod from_tensor(tensor, rank, factorization='CP', **kwargs)[source]¶

Create a factorized tensor by decomposing a dense tensor

Parameters:

tensortorch.tensor

tensor to factorize

rankint, ‘same’ or float

rank of the decomposition

factorization{‘CP’, ‘TT’, ‘Tucker’}, optional

Tensor factorization to use to decompose the tensor, by default ‘CP’

Returns:

TensorFactorization

Tensor in Factorized form.

Raises:

ValueError

If the factorization given does not exist.

forward(indices=None, **kwargs)[source]¶

To use a tensor factorization within a network, use tensor.forward, or, equivalently, tensor()

Parameters:

indicesint or tuple[int], optional

use to index the tensor during the forward pass, by default None

Returns:

TensorFactorization

tensor[indices]

property decomposition¶

Returns the factors and parameters composing the tensor in factorized form

to_tensor()[source]¶

Reconstruct the full tensor from its factorized form

dim()[source]¶

Order of the tensor

See also

ndim

Notes

fact_tensor.dim() == fact_tensor.ndim

property ndim¶

Order of the tensor

See also

dim

Notes

fact_tensor.dim() == fact_tensor.ndim

size(index=None)[source]¶

shape of the tensor

Parameters:

indexint, or tuple, default is None

if not None, returns tensor.shape[index]

See also

shape

normal_(mean=0, std=1)[source]¶

Inialize the factors of the factorization such that the reconstruction follows a Gaussian distribution

Parameters:

meanfloat, currently only 0 is supported

stdfloat

standard deviation

Returns:

self

property name¶

Factorization name (‘tucker’, ‘tt’, ‘cp’, …)