==========================
ADDED:
* Added `chaospy.orthogonal.frontend:generate_expansion` as an one stop
expansion generation function.
* Add tag-check when deploying using tags.
* Add logging which activates on env `CHAOSPY_DEBUG=1`.
Log to file with env `CHAOSPY_LOGFILE=/path/to/file`
* Added *Program Evaluation and Review Technique* (PERT) distribution.
* Adding support for `Dist.__matmul__`
(which obviously does nothing in python 2).
* Adding tests to the *hard-to-get-right* sub-module:
`chaospy.distributions.operators`.
* Added LRU cache to some quadrature schemes.
* Added segments to Newton-Cotes, Fejer and Clenshaw-Curtis
(as this is recommended to have to discretized Stieltjes).
* Added experimental Jupyter notebooks with user tutorials/recipes
`GITROOT/tutorial`
* Gumbel and Clayton copulas get analytical recursive Rosenblatt
transformations.
CHANGED:
* Update `numpoly` to version 0.3.0.
* Replace explicit numpoly import, with an implicit one with a
smart-wrapper.
* Docs updated with new polynomial string representation order.
* Update to documentation.
* Replace sample and quadrature scheme name from one letter
["G", "E", "C", "H", ...], to new full name strings:
["gaussian", "legendre", "clenshaw-curtis", "halton", ...].
(Old style still works, but is undocumented.)
* Increase quadrature sample rate 100->200 when doing discretized Stieltjes
to increase accuracy (at the computational cost).
* Increased sample rate for approximate inverse (used when inverse is
missing), increasing accuracy at extra computational cost.
* New style Archemedean copula.
* Refactor `chaospy.distributions.operators` to become less messy.
* Some adjustment to the expansion functions to align with the new frontend.
* Update lagrange to use `numpoly.bindex` in the backend.
* Use `graded: bool` and `reverse: bool` as a replacement for `sort: str =
"GRI"`:
* The `"I"` in `"GRI"` is deprecated: It can always be achieved with
`values = values[::-1]`, so it serves little purpose.
* The `"R"` was implemented backwards. `R` present is equivalent with
`reverse=False`.
* `sort` still works, but raises an warning about future deprecation.
* Using one letter strings is less readable, and needs to be removed.
Splitting them up, simplifies documentation.
REMOVED:
* Deprecating copulas Frank, Joe and Ali-Mikhail-Haw, as their accuracy is
not good enough.
* Remove really old tutorial stuff not longer in use.