- 2021-04-12
- MAJOR BUG FIX with smoothed latent state covariance
- 2019-10-04
- Multiple imputation for missing data with dynr.mi() function
- New demo for multiple imputation called MILinearDiscrete
- New demo for time-varying parameters called SDETVP
- New wrapper functions for computing smoothed derivative estimates using penalized B-splines implemented with the fda package
- New plotting functions, including functions to generate diagnostic plots from smoothed derivative estimates, and plot phase portraits
- New demo for computing and visualizing smoothed derivative estimates in GetDerivs
- New functionality in dynr.cook() to estimate continuous-time dynamic models with mixed effects through use of theta.formula
- MAJOR BUG FIXES with missing data
- 2019-04-01
- Outlier detection with dynr.taste() function
- Oulier removal and re-fitting with dynr.tast2() function
- New demo for outliers called OutlierDetection
- We now allow 1-regime recipe parts to co-exist with n-regime parts
- Lots of error checking was added around matching the number of regimes
- Many cases of the doDykstra error are now safely caught
- Generally cleaned up the error handling on models that failed to converge
- Shorten several demos to run faster
- 2018-09-21
- You don't need to install R on Windows to C:/R anymore! The default (C:/Program Files/R) now works.
- New demo for time-varying parameter (TVP) models
- Several new vignettes covering a range of topics
- The deviation form of regimes now displays properly in plotFormula
- No longer require 'outfile' specification in dynr.model()
- Fix a pointer addressing issue that could have caused crashes
- 2018-02-08
- New 'verbose' argument to dynr.cook turns off printing of optimization history
- New demo for Process Factor Analysis (PFA)
- Regime-switching printing in plotFormula() with new 'printRS' argument
- Greatly improved convergence rates for all models
- Allow full initial covariance estimation
- Fixed major bug in regime-switching matrix dynamics that formerly crashed R
- 2017-08-21
- Noise printing by plotFormula() function
- Fixed innovation vector computation for larger than 1-dimensional observations
- 2017-06-16
- New demo for a linear oscillator with time-varying parameters
- Fixed printex output for covariates and deviation form of the initial conditions
- Fixed memory leak for intercepts in measurement models
- 2017-05-19
- Use of individual-level covariates in the initial conditions. See ?prep.initial for details.
- Deviation form of regime-switching models. Seee ?prep.regimes for details.
- Access to the predicted, filtered, and smoothed latent variable estimates, and other by-products from the regime-switching extended Kalman filter in the 'cooked' model.
- We now allow calculation of the negative log-likelihood value, the hessian matrix, and the predicted, filtered, and smoothed latent variable estimates at fixed parameter values without parameter optimization.
- Beta version of a multiple imputation procedure. See ?dynr.mi for details.
- Fixed a rounding bug that improves free parameter optimization, especially for models with many observed variables.
- Improved documentation throughout
- Added more examples in the help pages
- 2017-02-21
- A new demo example is added to replicate the results from Yang & Chow (2010) paper.
- Some standard S3 methods are added for the dynrCook class object.
- autoplot() is added as an alias for dynr.ggplot().
- dynr.data() now automatically handles ts class objects and equally spaced data with missingness.
- Changes are made to accommodate the new release of ggplot2.
- 2016-08-12
- In single-regime models, free parameters for intercepts and covariate effects in the measurement model can now be properly estimated.
- Standard errors are more frequently returned
- Flags indicate problematic standard errors.
- Warning messages are more helpful regarding standard errors.
- A weight flag allows easier convergence of multi-subject models.
- Several new plotting features.
- 2016-06-07
- Initial release to CRAN!