* patch to column slicers.
* Substantial speed up in triangle initialization for larger triangles
* Fancy loc/iloc indexing now honors the order the user chooses
* Mack Std Err for tail fix xonsistent with R v0.2.11 fix
* Valuation triangles should now be JSON serializable
* Bug in expected loss method `predict` that wasn't rendering prediction results correctly
* Full GPU-support using `cupy`.
* quick patch for `Pipeline`
* Fixed an inappropriate mutation in the `Benktander` estimator when calls to `predict` are made
* Allow for `sample_weight` argument to be directly passed to a `Pipeline`
* Added `broadcast_axis` for better `BootstrapODPSample` support
* Eliminated a side-effect in `Development` when using `drop`
* Improved dropping capabilities of `BootstrapODPSample`
* Fixed TailCurve shape misaligned 57
* Fixed triangle json serializers to include all data necessary to recreate triangle
* Fixed various edge case issues when `len(development) == 1`
* Fixed `MackChainladder` assymentric triangle support (e.g. `OYDQ`)
* Fixed `MunichAdjustment` assymentric triangle support (e.g. `OYDQ`)
* Fixed datetime flexibility on dataframe ingestion into triangle
* aligned `append` method with pandas
* Moved excel exhibit functionality to a separate [xlcomose](https://xlcompose.readthedocs.io/en/latest/index.html) package.
* patched a defect in `Triangle.grain()`
* patched a defect in exhibits.py pertaining to titles
* Fixed issues in asymmetric triangles
* Replaced `Exhibit` API with `Row`, `Column`, `Tabs`, and `DataFrame`
Minor release to improved commutative properties of triangle functions (`grain`, `val_to_dev`, `dev_to_val`, `cum_to_incr`, `incr_to_cum`, `latest_diagonal`) for all types of triangles (development triangles, valuation triangles, complerte triangles)
* patch to asymmetric LDF fix in release 0.4.0
* Added `Triangle.dropna()` method that allows shaving off origins/development vectors that are all 0/NA
* Added better datetime management when instantiating a triangle
* Added json serializers for `Triangle` as well as the estimators and `Pipeline`. These can be accessed with the `to_json` method and `cl.read_json`
* Added `DevelopmentConstant` estimator that takes LDF or CDF patterns as a dictionary instead of calculating from a triangle. This is useful for incorporating industry patterns into an analysis
* General improved functionality for malformed triangles
* improved memory management by eliminating unnecessary usage of `deepcopy`
* Fixed an issue where LDFs were calculated incorrectly for asymmetric triangles
* Fixed a bug where slicers weren't working properly after particular mutations (e.g. adding a new column)
* Arithmetic on triangles with different origin ranges will now take the union of the origin periods. Previously the intersection was taken.
* Added predict functionality IBNR methods
* Fixed mutation in `TriangleGroupBy` aggregate methods
* Fixed bug in `Triangle.grain` method
* Fixed issue 43 User-Specified columns should be honored
* Fixed issue 44 cdf_ labels should be 'xx-Ult'
* Fixed issue 45 MackChainladder.total_mack_std_error_ TypeError
* Added ability `to_pickle` and `read_pickle` functionality to Triangles and Estimators.
* Minor bug in valuation slicing that restated the valuation_date of triangle incorrectly.
* 34 added a decay period to `TailConstant` to allow for run-off patterns beyond the end of the triangle for Actual vs. Expected analysis
* 41 Added Triangle.dev_to_val() and Triangle.val_to_dev() to allow switching between a left-aligned (development) and right-aligned (valuation) Triangle
* 40 Fixed bug that did not replace column in `Triangle` when assigned to existing column name
* `Triangle.grain` works with incremental and cumulative triangles now
* `Triangle.trend` has been changed to trend along valuation period
* Added substantial functionality to `Development` allowing end user to omit any specified link ratios.
* Converted `origin` and `valuation` from `DateTimeIndex` to `PeriodIndex` so that they work better with pandas datetime functionality.
* 'regression' and 'simple' averages were swapped in `Development`, this has been corrected
* Fixed bug that didn't allow `TailCurve` to be fit directly to a `Triangle`.
* Refactored `WeightedRegression` class to be in `sklearn` style
* Added `Triangle.valuation` accessor to be used similar to `origin` and `development`. This allows for slicing valuations (e.g. diagonals).
* Made triangle arithmetic more robust.
* Extended tail patterns from a point estimate to a one year run-off plus a point estimate. This is to facilitate Actual Vs Expected analysis for the year following an analysis.
* Created `Triangle.values` property to align with pandas-style syntax.
* Backward compatible to python 3.5
* Created `BootstrapODPSample` class to perform ODP Bootstrap sampling of triangles
* Added more aggregate functions to `Triangle` including: `mean`, `median`, `max`, `min`, `prod`, `var`, `std`
* Added several more pandas passthrough methods including: `to_dict`, `unstack`, `pivot`, `drop_duplicates`, `describe`, `melt`
* Added functionality to the `Triangle.to_frame` method to allow any 4D Triangle to be recast as a `DataFrame` as long as any two of its axes are of length=1.
* Bug fix in `Gridsearch` that didn't allow for passing of a `sample_weight` when used in conjunction with `Pipeline`
* Altered `Triangle.rename` method to be more consistent with the pandas implementation
* hotfix to `CapeCod` to make it work properly
* CapeCod Unit Test
* added `sample_weight` to `GridSearch.fit()`
* Lot's o' documentation
* converted sample datasets to `csv` format from `pkl` format
* Modifed core.Triangle to use parameter `index` in place of `keys` and `columns` in place of `values`. This is done to promote consistency with the pandas API as well as allowing for `values` property to be used in extended functionality of the Triangle
* Added `warnings` to development.Development to warn on failure of Mack Standard Error when only one period is used in the estimation of LDFs
* Added workflow.Pipelne and workflow.GridSearch for scikit-learnesque scenario testing.
Significantly overhauled API to be more consistent with pandas as scikit-learn.
Added Munich Chainladder functionality.