# 2025-02-20 v.1.0.0 [AG]
- Rewrite of tsaux for release to CRAN (as it is required by upcoming re-write of tsissm).
- Added vignette
- Added unit tests

# 2022-04-30 v.0.3.0 [AG]
- Added missing hanlding for some additional metrics (weighted matrix based)
# 2022-03-17 v.0.2.0 [AG]
- Added additional checks to auto_regressors (and auto_clean) for frequency
and data size (for seasonal) to avoid errors in stlm when length(y) <= 2xfrequency
# 2022-02-25 v.0.1.9 [AG]
- Added a general tstransform function for capturing all the transformations.
Added ldots to the functions to simplify calling from other functions.
and returning a list with the transform and inverse functions
# 2022-02-17 v.0.1.8 [AG]
- Added the logit transformation for bounded outcomes
# 2021-12-01 v0.1.7 [AG]
- Added the auto_clean function to automatically clean outliers or temporary
changes from a series.
# 2021-12-01 v0.1.6 [AG]
- Added the smape, rmape and msis metrics. The rmape is the rescaled measure
of MAPE based on a Box Cox transform from the paper by Swanson et al (2011).
# 2021-08-05 v0.1.5 [AG]
- Reversed changes to v0.1.4 for auto_regressors and re-implemented approach
using tso with initial outlier detection, followed by stl decomposition and
then tso on the deseasonalized series. The default options should be carefully
examined prior to use, as this remains an experimental function which may be
changed in the future.
# 2021-08-01 v0.1.4 [AG]
- Added the UC as an option in the auto_regressors function, based on the
UComp package. This is much faster and more stable than the tso function.
# 2021-07-08 v0.1.3 [AG]
- Added the auto_regressors function to the package for automatic detection
of outliers, trend breaks and temporary changes (wrapper for tso function of
tsoutliers package).
# 2021-04-20 v0.1.2 [AG]
- Added ability to have missing values in tslinear (and predict missingness) as
a requirements for implementing ETS with missing values (tslinear provides
starting values for the states).
- Replaced stl with stlplus to allow missing values for the multivariate
Box Cox transformation (which required deasonalized values).
- Added more checks for missing values in metrics.
- ToDo: add more checks in the metrics (e.g. crps) and weighted metrics