- Fixing errors on the
`train_model`

function:- Error with the forecast output
- Error with the nnetar model

- Replacing the
`xts::indexClass`

function with`xts::tclass`

function - Removing the
`ts_backtesting`

function, which was replaced by the`train_model`

function - Removing the
`ts_acf`

and`ts_pacf`

functions, the`ts_cor`

will replace them - Removing the
`bsts`

package from the package dependency

Package license

- Changing the package license from GPL-3 to MIT

New functions

*train_model - a flexible framework for training, testing, evaluating, and forecasting models. This function provides the ability to run multiple models with backtesting or single training/testing partitions * plot_model - animation the performance of the train_model output on the backtesting partitions * plot_error - plotting the error distribution of the train_model output * ts_cor - for acf and pacf plots with seasonal lags * arima_diag - a diagnostic plot for identify the AR, MA and differencing components of the ARIMA model

Deprecated functions

- ts_backtesting - will be replaced by the train_model function
- ts_acf / ts_pacf functions - will be replaced by the ts_cor function

Fix errors * ts_seasonal - aligning the box plot color * ts_plot - setting the dash and marker mode for multiple time series

New functions * forecast_sim - creating different forecast paths for forecast objects (when applicable), by utilizing the underline model distribution with the simulate function * ts_grid - tuning time series models with grid search approach using backtesting method. Currently, support only the Holt-Winters model * plot_grid - plotting the output of the ts_grid function

Fix errors * ts_plot, test_forecast - avoid default setting of the plot_ly function, and set explicitly the plot setting (e.g., color, line mode, etc.). This allows using the function with the plotly subplot function * ts_seasonal - define the order of the frequency units of the box plot option plot_forecast - fixing a gap between the forecast values and the time (x-axis) values

- ts_to_prophet function for converting ts objects (“ts”, “zoo” and “xts” class) to prophet object
- ccf_plot function for plotting corss correlation lags between two time series
- Fixed error in the ts_backtesting function - supprting xreg option

New functions: * ts_backtesting - a horce race of multiple forecasting models with backtesting * ts_quantile - time series quantile plot for time series data * ts_seasonal - supports multiple inputs and new color palattes

What’s new: * New options for the seasonality plot * Heatmap and surface plots * Polar plot * Converting function from xts and zoo to ts class * Spliting function for ts object for training and testing partitions

- Time series lags plot - ts_lags() function
- Function ts_split() to split ‘ts’ object into training and testing partitions
- Functions for converting xts and zoo objects for ts object:
- xts_to_ts(), and
- zoo_to_ts()

- Two types for the seasonal_ly() plot:
- “normal” - seasonal variation by year, or
- “cycle” - seasonal variation by the cycle units over time (months or quarters)
- “polar” - polar plot for seasonality
- “box” - box-plot by cycle units

- Decompose plot with the decompose_ly() function

- Data set - US monthly total vehicle sales: 1976 - 2017 (USVSales), ‘ts’ object
- Data set - US monthly civilian unemployment rate: 1948 - 2017 (USUnRate), ‘ts’ object
- Data set - US monthly natural gas consumption: 2000 - 2017 (USgas), ‘ts’ object
- Data set - University of Michigan Consumer Survey, Index of Consumer Sentiment: 1980 - 2017 (Michigan_CS), ‘xts’ object
- Data set - Monthly crude Oil Prices: Brent - Europe: 1987 - 2017 (EURO_Brent), ‘zoo’ object

- Function for plotting univariate and multivariate time series data
- Evaluation plot for the testing set (hold-out data)
- Interactive seasonality plot
- Functions for interactive plot for the ACF and PACF