Improved initialization of the numerical likelihood optimization.

Now the states after model estimation are automatically ordered according to the estimated mean of the state-dependent distributions, see

`reorder_states()`

with the new (default) option`state_order = "mean"`

.Re-fitted the example models contained in the package.

Added examples to

`fit_model()`

.Small code improvements in file

`ll.cpp`

.

- Small bug fix when computing the stationary distribution.

Controls can now be provided separately for the

`set_controls()`

function.The arguments in

`fHMM_parameters()`

for model parameters were slightly renamed as follows:`mus`

->`mu`

`sigmas`

->`sigma`

`dfs`

->`df`

`Gammas_star`

->`Gamma_star`

`mus_star`

->`mu_star`

`sigmas_star`

->`sigma_star`

`dfs_star`

->`df_star`

The log-normal state-dependent distribution is renamed:

`lnorm`

->`lognormal`

.Two more state-dependent distributions were added:

`normal`

and`poisson`

.The Viterbi algorithm can be directly accessed via

`viterbi()`

.Renamed

`simulate_data()`

->`simulate_hmm()`

to make the functionality clearer. Furthermore, this function is now exported and can be used outside of the package to simulate HMM data.`download_data()`

no longer saves a .csv-file but returns the data as a`data.frame`

. Its`verbose`

argument is removed because the function no longer prints any messages.The utilities (i.e., all functions with roxygen tag

`@keywords utils`

) were moved to the`{oeli}`

package.

- Fixed documenting the new special sentinel "_PACKAGE" for the package help file, see https://github.com/r-lib/roxygen2/issues/1491.

Extended the time horizon of saved data and updated models for demonstration.

The

`download_data()`

function now returns the data as a`data.frame`

by default. However, specifying argument`file`

still allows for saving the data as a .csv file.The

`plot.fHMM_model()`

function now has the additional argument`ll_relative`

(default is`TRUE`

) to plot the relative log-likelihood values when`plot_type = "ll"`

.Significantly increased the test coverage and fixed minor bugs.

Changed color of time series plot from

`"lightgray"`

to`"black"`

for better readability.Added a title to the time series plot when calling

`plot.fHMM_model(plot_type = "ts")`

. Additionally, a time interval with arguments`from`

and`to`

can be selected to zoom into the data.

Added the following methods for an

`fHMM_model`

object:`AIC()`

,`BIC()`

,`logLik()`

,`nobs()`

,`npar()`

,`residuals()`

.The log-normal distribution can now be estimated by setting

`sdds = "lnorm"`

in the`controls`

object.

Fixed bug in

`reorder_states()`

that did not order the fine-scale parameter sets when the coarse-scale order was changed.Fixed bug in

`parameter_labels()`

that returned the wrong order of parameter labels.Changed plot type of simulated data to lines.

In the vignette on controls, in the section about example specifications for

`controls`

, corrected`sdds = "gamma(mu = -1|1)"`

to`sdds = "gamma(mu = 0.5|2)"`

because mean of the Gamma distribution must be positive.Added

`digits`

argument to`print.fHMM_predict()`

.Fixed bug in

`reorder_states()`

that allowed for misspecification of`state_order`

.Added option to

`fit_model()`

to initialize at the estimates of another model (#73).

Enhanced the package by S3 classes.

Added more

`controls`

specifications.Included a prediction function.

Improved documentations.

Added vignettes.

Improved specification of

`controls`

.Fixed minor bugs.

Improved documentation of functions and README.

Improved specification of

`controls`

. (#37 and #38)

- Initial version.