Equadratures

New additions include:

- Bayesian polynomials
- Graphical polynomial models
- Logistic polynomial regression over manifolds

This new minor version (9.1.0) contains a number of new features:

- **[Datasets](https://equadratures.org/_documentation/datasets.html) module:** This module has been consolidated and enhanced. It now contains numerous utility tools for pre and post-processing of data, as well as a method to load example datasets from our [datasets repo](https://github.com/Effective-Quadratures/data-sets) (see [this post](https://discourse.equadratures.org/t/datasets-module/125))This text will be hidden.
- **[Scaler](https://equadratures.org/_documentation/scalers.html) classes:** Standardisation/normalisation of data can now be carried out with our scaler classes. These make it easier to transform train and test splits with the same transformation, and untransform data as required.
- **[PolyTree](https://equadratures.org/_documentation/polytree.html) module:** The polynomial regression tree module has been significantly enhanced, with new splitting criteria, utility methods, and plotting functionality.
- **[Solver](https://equadratures.org/_documentation/solver.html) class:** The underlying solvers used by *equadratures* polynomials have been reworked into sub-classes of the Solver class. This allows for utility methods to be attached to the various solvers, as well as standalone testing of solvers, and the addition of a custom solver wrapper for prototyping purposes (see [this post](https://discourse.equadratures.org/t/solver-subclasses/121)).
- **[Plotting](https://equadratures.org/_documentation/plot.html) utilities:** A large number of plotting methods have been included, allowing users to rapidly generate plots from the various *equadratures* classes (see this [post](https://discourse.equadratures.org/t/in-built-plotting-methods/103)).
- **Docs:** The docs have been substantially updated and reworked.

Numerous minor fixes and enhancements have also been made. The core functionality of v9.1.0 should be backward compatible with the previous version (v9.0.1).

The `PolyTree` module has been updated to include:
- A `model_agnostic` tree induction algorithm based on the ideas of M5P model trees. This method is significantly faster than the `model_aware` method for larger datasets.
- Options for a complexity penalisation term and smoothing of predictions, both of which help prevent overfitting to training data.
- A post hoc pruning method to prune previously fitted trees. Pruning with held-out data can further reduce overfitting.

Release updates include:
- New Gram Schmidt based method for generating polynomial approximations over correlated spaces.
- Polynomial variance calculation based on uncertainty in data.
- Piecewise polynomial approximations via regression trees.
- Ridge approximation-based trust region optimization.
- Elastic net, Huber regression, least absolute residual, and sparse relevance vector machine methods for coefficient computation.
- Analytically defined weight functions (including piecewise ones) for generating orthogonal polynomials.
- Generation of built-in data-sets.