Hottbox

Package status on branch `develop`
[](https://travis-ci.org/hottbox/hottbox/)
[](https://ci.appveyor.com/project/IlyaKisil/hottbox-6jq6a/branch/develop)
[](https://coveralls.io/github/hottbox/hottbox?branch=develop)

Added
- [ ]

Changed
- [ ]

Removed
- [ ]

Fixed
- [ ]

Package status on branch `develop`
[](https://travis-ci.org/hottbox/hottbox/)
[](https://ci.appveyor.com/project/IlyaKisil/hottbox-6jq6a/branch/develop)
[](https://coveralls.io/github/hottbox/hottbox?branch=develop)

Changed
- [x] Copy button in docs puts content into clipboard
- [x] `Makefile` utilises `pipenv`
- [x] Better structure of `Makefile`

Fixed
- [x] Compatibility with `pandas==1.0.0`. Deprecation of `as_matrix()` in favour of `values`
- [x] Compatibility with future versions of `numpy`. Deprecation of non-tuple sequence for multidimensional indexing.

Added
- [x] `CMTF` - An algorithm for Coupled Matrix and Tensor factorization for two `Tensors` of order n and 2 with respect to a specified `rank`
- [x] `RandomisedCPD` - An algorithm for Randomised Canonical Polyadic Decomposition.
- [x] `Parafac2` - An algorithm for PARAFAC2 model computed via ALS
- [x] `TelVI` and `TelVAC` - Algorithms under Tensor Ensemble Learning (TEL) framework
- [x] Various utils for generations and validation of multi-dimensional arrays
- [x] Toggle button into code examples in API documentation. It hides python prompts and output, making copying samples much easier.

Changed
- [x] New look of our documentation page which is now based on [`guzzle_sphinx_theme`](https://github.com/guzzle/guzzle_sphinx_theme)
- [x] New structure of documentation, each method or class resides in a separate webpage (thanks to sphinx templating).

Added
- [x] Custom Exceptions that are more appropriate for the structure of `hottbox`
- [x] LS-STM (least squares support tensor machine) for classification
- [x] Auto deployment of the documentation that reflects source code at the `develop` branch
- [x] Source files and utils to that make setup of development environment easier

Added
- [x] Tools to convert multi-index pandas dataframe into a `Tensor` and vise versa.
- [x] Quick construction of generic objects of `Tensor`, `TensorCPD`, `TensorTKD` and `TensorTT` classes.
- [x] Class `Mode` for meta information about data modes of for tensor representations. It is stored in `_modes` as list.
- [x] Class `State` that tracks data manipulation operation applied to `Tensor`. It is stored in `_state`.
- [x] Option for creating a `Tensor` in the unfolded form.
- [x] Methods for (re)setting mode names and the corresponding indices for `Tensor`
- [x] Mode description (and the corresponding methods) for `TensorCPD`, `TensorTKD` and `TensorTT` classes by analogy with the `Tensor` class
- [x] Parameter `keep_meta` to `decompose` methods for the cpd and tucker type decompositions. Based on its value, meta information of the modes of `tensor` to be decomposed can be extracted and assigned to the `TensorCPD` and `TensorTKD` respectively.
- [x] Direct summation and comparison of `Tensor` objects (redefined `__add__`, `__eq__`)
- [x] Direct summation and comparison of `TensorCPD` and `TensorTKD` object (redefined `__add__`, `__eq__`)
- [x] Defined `__str__` and `__repr__` for `Tensor`, `TensorCPD`, `TensorTKD` and `TensorTT`
- [x] Defined `__repr__` for tensor decomposition algorithms.
- [x] Kolda folding and unfolding
- [x] Vectorisation method for a `Tensor` class
- [x] Restrictions on methods `fold`, `unfold` and `mode_n_product` of `Tensor`. Whether they can be called is determined by the current state of the `Tensor` object.

Changed
- [x] Each mode of a `Tensor` there is characterised by a corresponding `Mode` object with meta information
- [x] Mode names for the `Tensor` constructor should be passed as list instead of OrderedDict. These names are used to create `Mode` objects which are stored in a list `Tensor._modes`
- [x] Property `reconstruct` of `TensorCPD`, `TensorTKD` and `TensorTT` classes is now a method (should have been in the first place). Also it take optional parameter `keep_mata` for extraction of meta information about modes
- [x] `describe` functionality is now implemented by `__str__`. Instead, `describe` provides some statistics of the `Tensor` by analogy with `pandas`

Removed
- [x] Parameter `ft_shape` from the `TensorTT` constructor
- [x] Parameter `ft_shape` and the corresponding attribute from the `Tensor` constructor.
- [x] Parameter `mode_description` from constructors for all tensor decomposition algorithms
- [x] Attribute `_mode_names` from the `Tensor`

Fixed
- [x] Fix copy methods for `TensorCPD` and `TensorTKD` due to new attributes

Added
- `copy` method for the core tensor structures
- `describe` method that describes an instance of `Tensor` 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` for `HOSVD` and `HOOI` classes