Implements Dirichlet regression models.
News for Package 'DirichletReg'
Changes in Version 0.6-2:
� Fixed an error in the ternary plot (bottom axis tick labels were printed in reversed order; pointed out by Emilio A. Laca).
� Imports functions from default packages, as required by the new CRAN-check. � Changed the 'Description' field to eliminate the note "Malformed Description field: should contain one or more complete sentences." when checking the package.
Changes in Version 0.6-1:
� Re-derived gradients/Hessians, optimized C-code, and tweaked computation of starting values which leads to considerably better performance.
� Fixed some bugs in 'drop1'.
� Expanded 'testthat' checks.
Changes in Version 0.6-0:
� All likelihood and gradient functions are now written in C (all called via '.Call()' now instead of '.C()') which leads to a considerable gain in speed.
� A 'drop1' method for Dirichlet regression models was added. As it is still experimental and will probably change, use it with care. Options such as 'scope' will be added in one of the next releases.
� Fixed a bug in the 'print' method for confidence intervals and one in 'confint.DirichletRegModel()'.
� Expanded the 'testthat' check- and test-suite.
� Known issue: If you have collinear (aliased) terms, the estimation will fail. This will be handled automatically in subsequent releases, but for now, please remove the respective terms. If you fit a model and it says something like:
Error in prepareFixed(start = start, activePar = NULL, fixed = fixed) : At least one parameter must not be fixed using argument 'fixed' you most likely have collinear terms or "empty" combinations of interaction terms.
Changes in Version 0.5-2:
� Fixed checking functions in 'tests/testthat'.
Changes in Version 0.5-1:
� Fixed a bug when using the 'subset' argument.
� Added tolerance for normalization check to 'DR_data'.
� moved 'NEWS' to the new fancy 'NEWS.Rd' file/format.
� Added the possibility to do quick analyses and transforming data "on the fly", like 'DirichReg(DR_data(A) ~ 1)'. However this is only intended for quick checking purposes and may be removed at any time.
Changes in Version 0.5-0:
� Transformation in 'DR_data' is now not only 'TRUE'/'FALSE', but, by default, a small numeric value to avoid troubles with floating point numbers close to 0 or 1.
� Time-critical routines were implemented in 'C' (pure R versions are available, see '?ddirichlet').
� 'anova.DirichletRegModel' now invisibly returns results as an object that is printed by a method.
� Optimized estimation routines.
� Fixed a bug in the predict method.
� Started development of a comprehensive test-suite using 'testthat'.
� Published a working paper on the package: Maier, M. J. (2014). DirichletReg: Dirichlet Regression for Compositional Data in R. Research Report Series / Department of Statistics and Mathematics, 125. WU Vienna University of Economics and Business, Vienna. <URL: http://epub.wu.ac.at/4077/>
� Added vignette with code to the working paper.
� Added citation info.
Changes in Version 0.4-1:
� The 'trafo' Argument of 'DR_data' has been changed, because it has lead to problems in practical applications when numbers very close to 0 or 1 were present.
� 'DR_data' checks for negative values and generates an appropriate error message.
� 'DR_data' has been made more robust in the presence of 'NA's.
Changes in Version 0.4-0:
� Data structure generated by 'DR_data' has changed � the new objects can now be integrated into data frames.
� Formula processing is now handled by the package 'Formula'.
� New methods have been implemented, especially for the class 'DirichletRegModel'.
� The documentation is now quite complete.
� Some speed improvements could be achieved.
� Lots of minor (invisible) changes.
Changes in Version 0.002:
� Added the analytical Gradient and Hessian for both parametrizations.
� Optimization: preliminary results by BFGS that become starting values for Newton-Raphson optimization computing the final results.
� Implemented some residuals
� Updated help entries