Changes in version 1.1.0                        

  - System C++ functions are now compiled on the first call to a method,
    and are only recompiled if the model equation changes.
  - The C++ functionality for 'filter', 'predict' and 'simulate' is now
    greatly improved:
      - The compilation time is reduced from 5-10 seconds to 2-3
        seconds.
      - The computations are faster.
      - The report functionality is now carried out in C++ (rather than
        pure R code), and is faster.
      - The three methods now also return forecasted observations, and
        associated dispersion.
  - The reporting for the 'estimate' method is now faster, due to
    reporting improvements (see above)
  - Previously the package returned data.frames as outputs in most
    cases. This has now been changed to matrices since these are usually
    faster to work with internally. Users must therefore use e.g. [,"t"]
    instead of $t when accessing columns.

                        Changes in version 1.0.2                        

  - Previously only 'ekf' had support for prediction and simulation. Now
    'lkf' and 'ukf' are also supported in both R and with Rcpp.
  - New method 'setAdvancedSettings' for various extra features
    (force.ad and tape configurations TMB::config and RTMB::TapeConfig)
  - New method 'setTrainingMethod' which enables training the likelihood
    as an ODE minimizing distance to full prediction - still need
    support for k-step.

                 Changes in version 1.0.1 (2025-08-27)                  

  - Internal large code-clean-up
  - Rcpp RNG simulations now use 'zigg' instead of 'RcppZiggurator'
    thereby avoiding need for 'SystemRequirements: GNU GSL'.
  - Removed RTMBode possibilities.
  - Renamed smoothing method 'laplace2' to 'laplace.thygesen'.
  - Fixed a simple typo bug in the results reporting functions.
  - Added an the implicit euler ODE solver to 'ode.solver' options.

                 Changes in version 1.0.0 (2025-04-08)                  

  - Initial release