Chaospy

==============================

Mv-KDE support!

ADDED:
* Added support for multivariate kernel density estimation distribution
`GaussianKDE`.
* Tutorial for KDE.
CHANGED:
* Default tolerance for the accuracy in approximate inverse lowered from
`10^-5` to `10^-12`.
* Lots of distribution have fixes such that `dist.inv([0, 1])` is now allowed
in general.
* Update to lots of docs to include example with `dist.inv([0, 1])`.
* Update `nbval` config to be more relaxed during tests.

==============================

Distribution operations are now all one-dimensional. One pass per dimension.

ADDED:
* New `report_on_error` decorator to get more understandable error output.
* New helper functions: `format_repr_kwargs`, `init_dependencies`,
`declare_dependencies`, `check_dependencies`.
* New intermediate distribution baseclasses:
`ItemDistribution`, `LowerUpperDistribution`, `MeanCovarianceDistribution`,
`OperatorDistribution`, `ShiftScaleDistribution`.
* New basic distribution: `InverseGamma`.
* New error type of error `UnsupportedFeatureError` to differentiate illegal
operations (covered by `StochasticallyDependentError`) and unsupported
features.
* Lots of new tests.
CHANGED:
* Lots and lots of positional `idx` arguments everywhere to indicate the
dimensions worked on. Except for `_mom` which is kept as is.
* Adding consistent baseclass naming convention:
`Copula{->Distribution}`, `Mul->Multiply`, `Neg->Negative`,
`DistributionCore->SimpleDistribution`.
* `Qoi_Dist` will no longer returns a numpy array in the multivariate case.
This is because `Distribution` no play will as a numpy object type.
* Changes to cache system:
* Cache content changed from `Dict[Distribution, ndarray]` to
`Dict[Distribution, Tuple[ndarray, ndarray]]` to store both inputs and
outputs for each calculations.
* backend function `_value` replaced with `_cache` for consistency.
* Backend interface `_get_value` replaced with `_get_cache_1` and
`_get_cache_2`. For former is new, the latter is a renaming.
* `Iid` is changed from being a function wrapper to a subclass wrapper,
allowing once again `isinstance(dist, Iid)`.
REMOVED:
* Deprecating topological soring in `J`, as this is now handled by the
evaluation order.
* Old function interfaces `add, mul, neg, trunk, trunc, pow`.
* Comparison operators `<`, `<=`, `>` and `=>` for distributions. These were
used as syntactic sugar referencing `chaospy.Trunc`. This to support `==`
operator.

===============================

Adding rotation: changing dist backend.

ADDED:
CHANGED:
* Baseclass distribution baseclass refactoring:
* Split old `Dist` into two: Abstract baseclass `Distribution` and
convenience structure `DistributionCore`.
* Cleaned up `__init__` structured to be more standardized.
* Much improved REPR handle.
* standardized `__len__`.
* Lots more pre-flight checks for distribution integrity.
* Simplification and standardization of `distributions.operators`.
* Better recursive caching of values during evaluations.
* Some hierarchy changes.
* Tiny changes in argument signature for some distribution. Same arguments,
but some change in names or order to standardize. These changes affect:
`Angelit`, `Burr`, `Cauchy`, `ChiSquared`, `F`, `FoldedNormal`,
`GeneralizedExtreme`, `HyperbolicSecant`, `Levy`, `LogWeibull`, `Logistic`,
`MvStudentT`, `Pareto1`, `Pareto2`, `PowerLogNormal`, `PowerNormal`,
`StudentT`,
REMOVED:
* `chaospy.distributions.evaluation` is removed in favor for method on the
`Distribution` class.
* `DependencyError` deprecated in favor of `StochasticallyDependentError`.
* `matmul` operator is in practice an really odd duckling that is highly
incompatible with the rotation idea. If linear map is needed, use
`MeanCovariance`.

===============================

Adding rotation: the fundamentals.

ADDED:
* Property for checking for dependencies: `Dist.stochastic_dependent`.
* Lots of illegal probability distribution configuration that would cause
trouble during execution are now caught earlier with an appropriate
error.
* Logging of samples out-of-bound for Dist methods:
`pdf`, `cdf`, `inv`, `fwd`.
* `Dist.pdf` get the extra flag `decompose` to split density into parts
(like `inv` and `fwd` does by default.) Should work with all
distribution, with a few exception. (MvLogNormal comes to mind.)
* New `LocScale` baseclass for all generic distributions with location and
and covariance structure.
* Lots of new tests.
CHANGED:
* New and improved dependency system based on underlying variable
declaration.
* Some probability distribution boundaries moved from hardcoded to
automatically detected.
* Update `Iid` to not be `J` subclass.
* Test cases for the new `LocScale` baseclass: `MvNormal` and `Alpha`.
REMOVED:
* Precedence order system. Was not ready yet, and a new one is being made
with the rotation system.

==========================

FIXED:
* Correct length for multivariate binary operators.

==========================

CHANGED:
* `numpoly` update to version 1.0.6.