Changelog • mpcmp

mpcmp 0.4.0

glm.cmp() now handles rank-deficient design matrices gracefully.
Fixed an error when using the gammastart argument (#11).
In the varying dispersion case, the coefficient table in the summary() output contained character columns instead of numeric ones; this has been fixed (#13).
Fixed a documentation issue where the cottonbolls dataset description was cut off due to an unescaped %.

autoplot() is now re-exported from ggplot2, and a print() method was added for cmp_autoplot_list objects.
Added a print.logLik.cmp() S3 method.
Updated ggplot2 code to use the linewidth aesthetic instead of the deprecated size for lines.
Renamed format.perc() to format_perc() to avoid clashing with an S3 method defined elsewhere.
Internal helper functions are no longer exported.
Fixed escaped asterisks in the AIC documentation formula.
Updated the package CITATION to use bibentry() format and secure URLs.
README refreshed: updated badges and switched installation instructions to pak.

Upgraded to testthat edition 3 and updated test syntax accordingly.
Added snapshot tests for summary(), influence(), hatvalues() and cooks.distance().
Adjusted tolerances in unit tests to account for numerical differences between operating systems.
Migrated to rhub v2, now run via GitHub Actions.

Added missing \value documentation to Rd files for exported functions, methods and datasets that lacked it, per CRAN feedback.
cmplrtest() now returns an object of class "htest" instead of printing results via cat(); the returned object is displayed via print.htest() and its statistic, degrees of freedom and p-value can be extracted directly.
LRTnu() now returns an object of class "htest" instead of printing results via cat(); the returned object is displayed via print.htest() and its statistic, degrees of freedom, p-value and log-likelihoods can be extracted directly.
plot.cmp() documentation example that was previously commented out is now runnable.

Exported more functions to NAMESPACE as requested by @yangchino1.
Patched a bug in predict.cmp that failed to handle new data with factors in it.

CRAN release: 2020-10-26

Added autoplot as an alias to gg_plot. Credit to Emi Tanaka (@emitanaka) for this suggestion.
Also patched an issue in the Rcpp code to pass cran check for solaris.
There are also other minor QOL improvements. # mpcmp 0.3.5
Added broom tidier methods support. Specifically added method for tidy(), glance() and augment().
summary() was rewritten in order to support tidier methods.
Also added method for vcov(), confint(), influence(), hatvalues(), rstandard(), cooks.distance().

Optimised rcomp() a bit by precalculating all the dcomp() values. Credit to Guilherme Parreira (@guilhermeparreira) for the issue request.
Documentations are now generated by Roxygen version 7.1.0.

Fixed an issue that offset term cannot be incorporated properly in the mean model. Credit to Sean Hardison (@seanhardison1) for finding this bug.

Updated glm.cmp() to allow varying dispersion. You can now link the dispersion parameters to some covariates via a log-link.
Most calculations are now performed inside fit_glm_cmp_const_nu() and fit_glm_cmp_vary_nu().
Functions such as print(), summary() are updated to support the updated glm.cmp().
Added spelltest as part of the testing procedure.
Added the sitophilus dataset to demonstrate the updated glm.cmp() function.

data(sitophilus)
M.sit <- glm.cmp(formula = ninsect ~ extract, formula_nu = ~extract, 
data = sitophilus)
summary(M.sit)

Added travis.CI support.
New functions gg_plot(), gg_histcompPIT() and gg_qqcompPIT() are added to provide the ggplots version of the diagnostic plots.
The package now depends on a more recent version of R ( $\geq$ 3.2).

Added a NEWS.md file to track changes to the package.
Added a draft logo to the package.
Ribeiro Jr et al. (2018) specification of the CMP model is utilised to provide a better initial estimate for the dispersion parameter. Added comp_mu_loglik_log_nu_only() to facilitate the optimisation.
Z(), the normalizing constant function, approximates its true value via (a fixed) truncation. This means the approximation would fail if the mean is large. The followings are implemented as a fix:
- A new function logZ() is created, based on a similar function in the cmpreg package of Ribeiro Jr, Zeviani & Demétrio (2019), and will supersede Z() due to its superior numerical stability.
- A Chebyshev’s inequality type argument is now implemented to have a more flexible upper truncation point.
glm.cmp(), dcomp(), pcomp(), qcomp(), rcomp() and functions that calculate expected values are updated to take advantage of these changes.

sum(0:500*dcomp(0:500,100,1.2))
sum(0:500*dcomp(0:500,150,0.8))
qcomp(0.6, 150, 1.2)

Added the fish dataset as a proof of concept that glm.cmp() can handle some larger count data.

data(fish)
M.fish <- glm.cmp(species~ 1+log(area), data=fish)
max(M.fish$fitted.values)

comp_lambdas() now has the ability to scale up & down lambdaub so that the correct $\lambda$ s can be found even if they are outside the preset boundary.

model.matrix() now retrieves the design matrix of the model properly.
glm.cmp() gains a few standard glm arguments: start, contrasts, na.action, subset.

CRAN release: 2019-03-04