TensorCPD¶

# Import statement

from hottbox.core import TensorCPD

class TensorCPD(fmat, core_values, mode_names=None)[source]¶

Representation of a tensor in the Kruskal form (CPD).

Parameters

fmatlist[np.ndarray]: List of factor matrices for the CP representation of a tensor
core_valuesnp.ndarray: Array of coefficients on the super-diagonal of a core for the CP representation of a tensor
mode_nameslist[str]: List of names for the factor matrices

Attributes

_fmat	(list[np.ndarray]) Placeholder for a list of factor matrices for the CP representation of a tensor
_core_values	(np.ndarray) Placeholder for an array of coefficients on the super-diagonal of a core for the CP representation of a tensor
_modes	(list[Mode]) Description of the factor matrix for the corresponding mode

Raises

TensorTopologyError: If there is inconsistency in shapes of factor matrices and core values

Examples

Create kruskal representation of a tensor with default meta information

>>> import numpy as np
>>> from hottbox.core import TensorCPD
>>> I, J, K = 5, 6, 7   # shape of the tensor in full form
>>> R = 4               # Kruskal rank
>>> A = np.ones((I, R))
>>> B = np.ones((J, R))
>>> C = np.ones((K, R))
>>> fmat = [A, B , C]
>>> core_values = np.arange(R)
>>> tensor_cpd = TensorCPD(fmat, core_values)
>>> print(tensor_cpd)
    Kruskal representation of a tensor with rank=(4,).
    Factor matrices represent properties: ['mode-0', 'mode-1', 'mode-2']
    With corresponding latent components described by (5, 6, 7) features respectively.

Create kruskal representation of a tensor with custom meta information

>>> import numpy as np
>>> from hottbox.core import TensorCPD
>>> I, J, K = 5, 6, 7   # shape of the tensor in full form
>>> R = 4               # Kruskal rank
>>> A = np.ones((I, R))
>>> B = np.ones((J, R))
>>> C = np.ones((K, R))
>>> fmat = [A, B , C]
>>> core_values = np.arange(R)
>>> mode_names = ["Year", "Month", "Day"]
>>> tensor_cpd = TensorCPD(fmat, core_values, mode_names)
>>> print(tensor_cpd)
    Kruskal representation of a tensor with rank=(4,).
    Factor matrices represent properties: ['Year', 'Month', 'Day']
    With corresponding latent components described by (5, 6, 7) features respectively.

Methods

`copy`(self)	Produces a copy of itself as a new object
`copy_modes`(self, tensor)	Copy modes meta from tensor
`reconstruct`(self[, keep_meta])	Converts the CP representation of a tensor into a full tensor
`reset_mode_index`(self[, mode])	Drop index for the specified mode number
`reset_mode_name`(self[, mode])	Set default name for the specified mode number
`set_mode_index`(self, mode_index)	Set index for specified mode
`set_mode_names`(self, mode_names)	Rename modes of a tensor representation

copy(self)[source]¶

Produces a copy of itself as a new object

Returns

new_objectTensorCPD

copy_modes(self, tensor)[source]¶

Copy modes meta from tensor

Parameters

tensor{Tensor, TensorCPD, TensorTKD, TensorTT}

Returns

selfTensorCPD

Notes

Most of the time this method should only be used by the decomposition algorithms

property core¶

Core tensor of the CP representation of a tensor

Returns

core_tensorTensor

property fmat¶

List of factor matrices for the CP representation of a tensor

Returns

factor_matriceslist[np.ndarray]

property ft_shape¶

Shape of a TensorCPD in the full format

Returns

full_shapetuple

property mode_names¶

Description of the physical modes for a TensorCPD

Returns

list[str]

property modes¶

Meta data for the factor matrices

Returns

list[Mode]

property order¶

Order of a tensor represented through the CPD

Returns

orderint

property rank¶

Rank of the CP representation of a tensor.

Returns

ranktuple

Notes

Most often referred to as the Kryskal rank

reconstruct(self, keep_meta=0)[source]¶

Converts the CP representation of a tensor into a full tensor

Parameters

keep_metaint: Keep meta information about modes of the given tensor. 0 - the output will have default values for the meta data 1 - keep only mode names 2 - keep mode names and indices

Returns

tensorTensor

reset_mode_index(self, mode=None)[source]¶

Drop index for the specified mode number

Parameters

modeint: Mode number which index to be dropped By default resets all indices

Returns

selfTensorCPD

reset_mode_name(self, mode=None)[source]¶

Set default name for the specified mode number

Parameters

modeint: Mode number which name to be set to default value By default resets names of all modes

Returns

selfTensorCPD

set_mode_index(self, mode_index)[source]¶

Set index for specified mode

Parameters

mode_indexdict: New indices for the factor matrices in form of a dictionary. Key defines the mode whose index to be changed. Value contains a list of new indices for this mode.

Returns

selfTensorCPD

set_mode_names(self, mode_names)[source]¶

Rename modes of a tensor representation

Parameters

mode_namesdict: New names for the tensor modes in form of a dictionary The name of the mode defined by the Key of the dict will be renamed to the corresponding Value

Returns

selfTensorCPD

Fork me on GitHub

Table Of Contents

TensorCPD¶