Releases: hottbox/hottbox
Releases · hottbox/hottbox
HOTTBOX [0.3.2] - Pandas and numpy compatibility
Changed
- Copy button in docs puts content into clipboard
-
Makefile
utilisespipenv
- Better structure of
Makefile
Fixed
- Compatibility with
pandas==1.0.0
. Deprecation ofas_matrix()
in favour ofvalues
- Compatibility with future versions of
numpy
. Deprecation of non-tuple sequence for multidimensional indexing.
HOTTBOX [0.3.1] - New look, New algorithms
Added
-
CMTF
- An algorithm for Coupled Matrix and Tensor factorization for twoTensors
of order n and 2 with respect to a specifiedrank
-
RandomisedCPD
- An algorithm for Randomised Canonical Polyadic Decomposition. -
Parafac2
- An algorithm for PARAFAC2 model computed via ALS -
TelVI
andTelVAC
- Algorithms under Tensor Ensemble Learning (TEL) framework - Various utils for generations and validation of multi-dimensional arrays
- Toggle button into code examples in API documentation. It hides python prompts and output, making copying samples much easier.
Changed
- New look of our documentation page which is now based on
guzzle_sphinx_theme
- New structure of documentation, each method or class resides in a separate webpage (thanks to sphinx templating).
HOTTBOX [0.2.1] - Introducing LS-STM (least squares support tensor machine) for classification
Added
- Custom Exceptions that are more appropriate for the structure of
hottbox
- LS-STM (least squares support tensor machine) for classification
- Auto deployment of the documentation that reflects source code at the
develop
branch - Source files and utils to that make setup of development environment easier
HOTTBOX [0.1.3] - Introducting tensor state and pandas integration tools
Added
- Tools to convert multi-index pandas dataframe into a
Tensor
and vise versa. - Quick construction of generic objects of
Tensor
,TensorCPD
,TensorTKD
andTensorTT
classes. - Class
Mode
for meta information about data modes of for tensor representations.
It is stored in_modes
as list. - Class
State
that tracks data manipulation operation applied toTensor
.
It is stored in_state
. - Option for creating a
Tensor
in the unfolded form. - Methods for (re)setting mode names and the corresponding indices for
Tensor
- Mode description (and the corresponding methods) for
TensorCPD
,TensorTKD
andTensorTT
classes
by analogy with theTensor
class - Parameter
keep_meta
todecompose
methods for the cpd and tucker type decompositions.
Based on its value, meta information of the modes oftensor
to be decomposed can be extracted
and assigned to theTensorCPD
andTensorTKD
respectively. - Direct summation and comparison of
Tensor
objects (redefined__add__
,__eq__
) - Direct summation and comparison of
TensorCPD
andTensorTKD
object (redefined__add__
,__eq__
) - Defined
__str__
and__repr__
forTensor
,TensorCPD
,TensorTKD
andTensorTT
- Defined
__repr__
for tensor decomposition algorithms. - Kolda folding and unfolding
- Vectorisation method for a
Tensor
class - Restrictions on methods
fold
,unfold
andmode_n_product
ofTensor
.
Whether they can be called is determined by the current state of theTensor
object.
Changed
- Each mode of a
Tensor
there is characterised by a correspondingMode
object with meta information - Mode names for the
Tensor
constructor should be passed as list instead of OrderedDict.
These names are used to createMode
objects which are stored in a listTensor._modes
- Property
reconstruct
ofTensorCPD
,TensorTKD
andTensorTT
classes is now a method
(should have been in the first place). Also it take optional parameterkeep_mata
for extraction
of meta information about modes -
describe
functionality is now implemented by__str__
. Instead,describe
provides some
statistics of theTensor
by analogy withpandas
Removed
- Parameter
ft_shape
from theTensorTT
constructor - Parameter
ft_shape
and the corresponding attribute from theTensor
constructor. - Parameter
mode_description
from constructors for all tensor decomposition algorithms - Attribute
_mode_names
from theTensor
Fixed
- Fix copy methods for
TensorCPD
andTensorTKD
due to new attributes
HOTTBOX [0.1.2] - CI setup and stability improvements
Added
copy
method for the core tensor structuresdescribe
method that describes an instance ofTensor
class- Mode descriptions for the modes of
Tensor
through the use of OrderedDict.
Modes can also be renamed - Input validation for constructors for
Tensor
,TensorCPD
,TensorTKD
,TensorTT
- Input validation for input data for
decompose
method for all tensor decomposition algorithms - Setup CI using Travis, AppVeyor and Coveralls
- Unit tests using pytest for all available modules
Changed
- Objects of
Tensor
,TensorCPD
,TensorTKD
,TensorTT
classes can only be created from numpy arrays - For all tensor representation all their data values can (should) only be accessed through corresponding properties.
- The original shape of the tensor can be defined during object creation of
Tensor
class super_diag_tensor
requires to pass a shape of desired tensor instead of its order
Fixed
reconstruct
was changing the original core so it was not possible to call it several times in a row- Incorrect size of a produced factor matrix when its computation is skipped in
decompose
forHOSVD
andHOOI
classes
HOTTBOX [0.1.1] - First release
Added
- Core operations of tensor algebra
- Classes for different tensor representations (
Tensor
,TensorCPD
,TensorTKD
,TensorTT
) - Functions for computing special types of tensors (
super_diag_tensor
,residual_tensor
) - Implementation of the most fundamental tensor decompositions (
CPD
,HOSVD
,HOOI
,TTSVD
) - Several methods for computing metrics of tensor decompositions
- Functions for estimating optimal Kryskal rank and computing multi-linear rank of a
Tensor