cran-comments.md
In nredell/forecastML: Time Series Forecasting with Machine Learning Methods

local Windows install, R 4.0.0 and 3.6.2
ubuntu 16.04 (on travis-ci), R 4.0.0 and 3.6.2

0 errors | 0 warnings | 0 notes

05-05-2020
This is a minor version update, 0.8.0 to 0.9.0, for an existing CRAN package.
- User-visible new features
  - New function create_skeleton() for analysis with big data.
  - New argument in create_lagged_df(..., predict_future = NULL) for predicting dynamic features.
  - Added the rmsse error metric from the M5 forecasting competition.
  - Forecast combination is clearer and now supports custom aggregation with combine_forecasts(..., aggregate = stats::median).
  - Added a new vignette for forecast combinations.
  - Continued improvement of plot aesthetics.
- Bug fixes
  - Multiple grouping columns now works with create_lagged_df().
  - Support outcome column positions other than 1 in create_lagged_df().
  - Plotting single time series predictions with non-contiguous windows removes connecting line.
- Internals
  - DESCRIPTION: works with dplyr < v1.0.0 and v1.0.0.
  - Reduced build time of existing vignettes.
  - Unit test code coverage = 83%.
02-28-2020
This is a minor version update, 0.7.0 to 0.8.0, for an existing CRAN package.
- User-visible new features
  - Support forecasting with multi-output models (e.g., multi-output neural networks).
  - Refined plot API including support for pivoting results with formulas.
  - Added a new vignette for forecasting grouped time series with factors.
- Bug fixes
  - Forecast combinations with grouped time series and factors now works.
  - Set required R version to 3.5.0
- Internals
  - Substantial code refactor
  - Unit test code coverage = 82%.
01-06-2020
This is a minor version update, 0.6.0 to 0.7.0, for an existing CRAN package.
- User-visible new features
  - Support forecasting factor outcome levels and probabilities.
  - Add function combine_forecasts() to produce final h-step-ahead forecasts from ensemble models.
  - Reworked vignettes for style and to cover the most common use cases.
- Bug fixes
  - None
- Internals
  - Refactored return_error() to support the addition of new error metrics in zzz.R
  - Unit test code coverage = 83%.
11-22-2019
This is a minor version update, 0.5.0 to 0.6.0, for an existing CRAN package.
- User-visible new features
  - Support sub-daily forecasting with the POSIXt datetime class.
  - Support a vector of custom validation windows in create_lagged_df() with window_start and window_stop.
  - The create_lagged_df(..., keep_rows = TRUE) argument allows retaining the first N rows of an ungrouped training dataset.
  - Simplified plots by removing window numbers from the legened with grouped time series.
  - Updated documentation.
- Bug fixes
  - MAPE error metrics were incorrectly set to median MAPE metrics for error results by window.
  - return_hyper() now returns the correct window number.
  - Printed forcats warning message suppressed when plotting empty data.frames which is the desired behavior.
  - Support plotting numerical and categorical hyperparameters in plot.forecast_model_hyper()
- Internals
  - Add dtplyr and data.table to the DESCRIPTION file and remove stringr.
  - Unit test code coverage increased from ~40% to 83%.
10-8-2019
This is a resubmission of a new-to-CRAN release. Changes include:
- Reduced DESCRIPTION from 103 to 74 words.
- Added a number of stop("message") checks on the input arguments for all functions and methods (~15 additional). The checks include various missing(), is.null(), and argument type checks.
- Minor changes in help docs for language consistency and to clarify required arguments and argument types.
- Added new unit tests; package has ~40% code coverage.
10-2-2019
- Reduced build time by several minutes (< 5 min w/ devtools::check())
- Removed donttest{} in all examples
- Fixed copyright format