Sequentia

Major changes

- Add support for dependent feature warping (addresses [124](https://github.com/eonu/sequentia/pull/124)). ([#135](https://github.com/eonu/sequentia/pull/135))
- Add multi-processed predictions for `HMMClassifier` (addresses [121](https://github.com/eonu/sequentia/pull/121)). ([#136](https://github.com/eonu/sequentia/pull/136))
- Re-order `predict()` and `evaluate()` arguments. ([138](https://github.com/eonu/sequentia/pull/138))

Minor changes

- Add `original_labels` documentation to `KNNClassifier`. ([133](https://github.com/eonu/sequentia/pull/133))
- Simplify `GMMHMM` documentation. ([134](https://github.com/eonu/sequentia/pull/134))
- Fix posterior comment in `classifier.svg`. ([137](https://github.com/eonu/sequentia/pull/137))

Minor changes

- Remove references to `sigment`. ([130](https://github.com/eonu/sequentia/pull/130))
- Fix type specifiers in documentation (see [129](https://github.com/eonu/sequentia/issues/129)). ([#131](https://github.com/eonu/sequentia/pull/131))

Major changes

- Switch out [`pomegranate`](https://github.com/jmschrei/pomegranate) HMM backend to [`hmmlearn`](https://github.com/hmmlearn/hmmlearn). ([#105](https://github.com/eonu/sequentia/pull/105))
- Remove separate HMM and GMM-HMM implementations – only keep a single GMM-HMM implementation (in the `GMMHMM` class) and treat multivariate Gaussian emission HMM as a special case of GMM-HMM. ([105](https://github.com/eonu/sequentia/pull/105))
- Support string and numeric labels by using label encodings (from [`sklearn.preprocessing.LabelEncoder`](https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.LabelEncoder.html)). ([#105](https://github.com/eonu/sequentia/pull/105))
- Add support for Python v3.6, v3.7, v3.8, v3.9 and remove support for v3.5. ([105](https://github.com/eonu/sequentia/pull/105))
- Switch from approximate DTW algorithm ([`fastdtw`](https://github.com/slaypni/fastdtw)) to exact implementation ([`dtaidistance`](https://github.com/wannesm/dtaidistance)) for `KNNClassifier`. ([#106](https://github.com/eonu/sequentia/pull/106))

Minor changes

- Switch to use duck-typing for iterables instead of requiring lists. ([105](https://github.com/eonu/sequentia/pull/105))
- Rename 'strict left-right' HMM topology to 'linear'. ([105](https://github.com/eonu/sequentia/pull/105))
- Switch `m2r` to `m2r2`, as `m2r` is no longer maintained. ([105](https://github.com/eonu/sequentia/pull/105))
- Change `covariance` to `covariance_type`, to match `hmmlearn`. ([105](https://github.com/eonu/sequentia/pull/105))
- Use `numpy.random.RandomState(seed=None)` as default instead of `numpy.random.RandomState(seed=0)`. ([105](https://github.com/eonu/sequentia/pull/105))
- Switch `KNNClassifier` serialization from HDF5 to pickling. ([106](https://github.com/eonu/sequentia/pull/106))
- Use [`intersphinx`](https://www.sphinx-doc.org/en/master/usage/extensions/intersphinx.html) for external documentation links, e.g. to `numpy`. ([#108](https://github.com/eonu/sequentia/pull/108))
- Change `MinMaxScale` bounds to floats. ([112](https://github.com/eonu/sequentia/pull/112))
- Add `__repr__` function to `GMMHMM`, `HMMClassifier` and `KNNClassifier`. ([120](https://github.com/eonu/sequentia/pull/120))
- Use feature-independent warping (DTWI). ([121](https://github.com/eonu/sequentia/pull/121))
- Ensure minimum Sakoe-Chiba band width is 1. ([126](https://github.com/eonu/sequentia/pull/126))

Major changes

- Stop referring to sequences as temporal, as non-temporal sequences can also be used. ([103](https://github.com/eonu/sequentia/pull/103))

Major changes

- Fix deserialization for `KNNClassifier`. ([93](https://github.com/eonu/sequentia/pull/93))
- Sort HDF5 keys before loading as `numpy.ndarray`s.
- Pass `weighting` function into deserialization constructor.

Major changes

- Fix `pomegranate` version to v0.12.0. ([79](https://github.com/eonu/sequentia/pull/79))
- Add serialization and deserialization support for all classifiers. ([80](https://github.com/eonu/sequentia/pull/80))
- `HMM`, `HMMClassifier`: Serialized in JSON format.
- `KNNClassifier`: Serialized in [HDF5](https://support.hdfgroup.org/HDF5/doc/H5.intro.html) format.
- Finish preprocessing documentation and tests. ([81](https://github.com/eonu/sequentia/pull/81))
- (_Internal_) Remove nested helper functions in `KNNClassifier.predict()`. ([84](https://github.com/eonu/sequentia/pull/84))
- Add strict left-right HMM topology. ([85](https://github.com/eonu/sequentia/pull/85))<br/>**Note**: This is the more traditional left-right HMM topology.
- Implement GMM-HMMs in the `GMMHMM` class. ([87](https://github.com/eonu/sequentia/pull/87))
- Implement custom, uniform and frequency-based HMM priors. ([88](https://github.com/eonu/sequentia/pull/88))
- Implement distance-weighted DTW-kNN predictions. ([90](https://github.com/eonu/sequentia/pull/90))
- Rename `DTWKNN` to `KNNClassifer`. ([91](https://github.com/eonu/sequentia/pull/91))

Minor changes

- (_Internal_) Simplify package imports. ([82](https://github.com/eonu/sequentia/pull/82))
- (_Internal_) Add `Validator.func()` for validating callables. ([90](https://github.com/eonu/sequentia/pull/90))