Include geometric Poisson Tweedie facilities.

Author: wbonat

DESCRIPTION

@@ -2,7 +2,7 @@ Package: mcglm
 Type: Package
 Title: Multivariate Covariance Generalized Linear Models
 Version: 0.5.0
-Date: 2019-03-19
+Date: 2019-09-10
 Author: Wagner Hugo Bonat [aut, cre],
   Walmes Marques Zeviani [ctb],
   Fernando de Pol Mayer [ctb]

R/mc_build_sigma.R

@@ -259,5 +259,81 @@ mc_build_sigma <- function(mu, Ntrial = 1, tau, power, Z, sparse,
             }
         }
     }
+    if(variance == "geom_tweedie") {
+        if (covariance == "identity" | covariance == "expm") {
+          Omega <- mc_build_omega(tau = tau, Z = Z,
+                                  covariance_link = covariance,
+                                  sparse = sparse)
+          V_sqrt <- mc_variance_function(
+            mu = mu$mu, power = power,
+            Ntrial = Ntrial, variance = "power", inverse = FALSE,
+            derivative_power = !power_fixed,
+            derivative_mu = compute_derivative_beta)
+          Sigma <- forceSymmetric(Diagonal(length(mu$mu), mu$mu^2) +
+                                    V_sqrt$V_sqrt %*%
+                                    Omega$Omega %*% V_sqrt$V_sqrt)
+          chol_Sigma <- chol(Sigma)
+          inv_chol_Sigma <- solve(chol_Sigma)
+          D_Sigma <- lapply(Omega$D_Omega, mc_sandwich,
+                            bord1 = V_sqrt$V_sqrt,
+                            bord2 = V_sqrt$V_sqrt)
+          if (power_fixed == FALSE) {
+            D_Sigma_power <- mc_sandwich_power(
+              middle = Omega$Omega,
+              bord1 = V_sqrt$V_sqrt, bord2 = V_sqrt$D_V_sqrt_p)
+            D_Sigma <- c(D_Sigma_power = D_Sigma_power,
+                         D_Sigma_tau = D_Sigma)
+          }
+          output <- list(Sigma_chol = chol_Sigma,
+                         Sigma_chol_inv = inv_chol_Sigma,
+                         D_Sigma = D_Sigma)
+          if (compute_derivative_beta == TRUE) {
+            D_Sigma_beta <- mc_derivative_sigma_beta(
+              D = mu$D,
+              D_V_sqrt_mu = V_sqrt$D_V_sqrt_mu, Omega$Omega,
+              V_sqrt = V_sqrt$V_sqrt, variance = variance)
+            output$D_Sigma_beta <- D_Sigma_beta
+          }
+        }
+        if (covariance == "inverse") {
+          inv_Omega <- mc_build_omega(tau = tau, Z = Z,
+                                      covariance_link = "inverse",
+                                      sparse = sparse)
+          Omega <- chol2inv(chol(inv_Omega$inv_Omega))
+          V_sqrt <- mc_variance_function(
+            mu = mu$mu, power = power,
+            Ntrial = Ntrial, variance = "power", inverse = FALSE,
+            derivative_power = !power_fixed,
+            derivative_mu = compute_derivative_beta)
+          D_Omega <- lapply(inv_Omega$D_inv_Omega,
+                            mc_sandwich_negative, bord1 = Omega,
+                            bord2 = Omega)
+          D_Sigma <- lapply(D_Omega, mc_sandwich,
+                            bord1 = V_sqrt$V_sqrt,
+                            bord2 = V_sqrt$V_sqrt)
+          Sigma <- forceSymmetric(Diagonal(length(mu$mu), mu$mu^2) +
+                                    V_sqrt$V_sqrt %*% Omega %*%
+                                    V_sqrt$V_sqrt)
+          chol_Sigma <- chol(Sigma)
+          inv_chol_Sigma <- solve(chol_Sigma)
+          if (power_fixed == FALSE) {
+            D_Sigma_p <- mc_sandwich_power(
+              middle = Omega,
+              bord1 = V_sqrt$V_sqrt,
+              bord2 = V_sqrt$D_V_sqrt_power)
+            D_Sigma <- c(D_Sigma_p, D_Sigma)
+          }
+          output <- list(Sigma_chol = chol_Sigma,
+                         Sigma_chol_inv = inv_chol_Sigma,
+                         D_Sigma = D_Sigma)
+          if (compute_derivative_beta == TRUE) {
+            D_Sigma_beta <- mc_derivative_sigma_beta(
+              D = mu$D,
+              D_V_sqrt_mu = V_sqrt$D_V_sqrt_mu, Omega = Omega,
+              V_sqrt = V_sqrt$V_sqrt, variance = variance)
+            output$D_Sigma_beta <- D_Sigma_beta
+          }
+        }
+    }
     return(output)
 }

R/mc_derivative_sigma_beta.R

@@ -35,5 +35,13 @@ mc_derivative_sigma_beta <- function(D, D_V_sqrt_mu, Omega, V_sqrt,
                                   bord2 = D_V_sqrt_beta)
         }
     }
+    if (variance == "geom_tweedie") {
+      for (i in 1:n_beta) {
+        D_V_sqrt_beta <- Diagonal(n_obs, D_V_sqrt_mu * D[, i])
+        output[[i]] <- Diagonal(n_obs, 2*D[, i]) +
+          mc_sandwich_power(middle = Omega, bord1 = V_sqrt,
+                            bord2 = D_V_sqrt_beta)
+      }
+    }
     return(output)
 }

R/mc_initial_values.R

@@ -93,7 +93,8 @@ mc_initial_values <- function(linear_pred, matrix_pred, link,
             }
         }
         if (variance[[i]] == "tweedie" |
-                variance[[i]] == "poisson_tweedie") {
+                variance[[i]] == "poisson_tweedie" |
+            variance[[i]] == "geom_tweedie") {
             power_initial[[i]] <- 1
             if (!is.null(offset[[i]])) {
                 data_temp <- data

R/mc_variance_function.R

@@ -37,7 +37,7 @@
 #'
 #' @source Bonat, W. H. and Jorgensen, B. (2016) Multivariate
 #'     covariance generalized linear models.
-#'     Journal of Royal Statistical Society - Series C X(X):XX--XX.
+#'     Journal of Royal Statistical Society - Series C 65:649--675.
 #'
 #' @export
 #'

README.Rmd

@@ -21,7 +21,8 @@ The `mcglm` package fits multivariate covariance generalized linear models
 
 ## Introduction
 
-`mcglm` fits multivariate covariance generalized linear models. It allows use a different linear predictor for each response variable of a
+`mcglm` fits multivariate covariance generalized linear models. 
+It allows use a different linear predictor for each response variable of a
 multivariate response. The response variable can be continous or
 dicrete, like counts and binary and also limited continuos ou
 discrete/continuous inflated responses. The most important and relevant

README.md

@@ -1,20 +1,20 @@
 
 
-# mcglm 0.4.0
+# mcglm 0.5.0
 
 [![Build Status](https://travis-ci.org/wbonat/mcglm.svg?branch=master)](https://travis-ci.org/wbonat/mcglm)
 Build status for the stable version (`master` branch)
 
 [![Build Status](https://travis-ci.org/wbonat/mcglm.svg?branch=devel)](https://travis-ci.org/wbonat/mcglm)
 Build status for the development version (`devel` branch)
 
-The `mcglm` package fit multivariate covariance generalized linear models
-(Bonat and Jorgensen, 2015).
+The `mcglm` package fits multivariate covariance generalized linear models
+(Bonat and Jorgensen, 2016).
 
 ## Introduction
 
-`mcglm` fit multivariate covariance generalized linear models. It allows
-use a different linear predictor for each response variable of a
+`mcglm` fits multivariate covariance generalized linear models. 
+It allows use a different linear predictor for each response variable of a
 multivariate response. The response variable can be continous or
 dicrete, like counts and binary and also limited continuos ou
 discrete/continuous inflated responses. The most important and relevant
@@ -37,20 +37,20 @@ library(devtools)
 install_github("wbonat/mcglm")
 ```
 
-Alternatively, download the package tarball: [mcglm_0.4.0.tar.gz][]
+Alternatively, download the package tarball: [mcglm_0.5.0.tar.gz][]
 and run from a UNIX terminal (make sure you are on the container file
 directory):
 
 
 ```
-R CMD INSTALL -l /path/to/your/R/library mcglm_0.4.0.tar.gz
+R CMD INSTALL -l /path/to/your/R/library mcglm_0.5.0.tar.gz
 ```
 
 Or, inside an `R` session:
 
 
 ```
-install.packages("mcglm_0.4.0.tar.gz", repos = NULL,
+install.packages("mcglm_0.5.0.tar.gz", repos = NULL,
                  lib.loc = "/path/to/your/R/library",
                  dependencies = TRUE)
 ```
@@ -62,13 +62,13 @@ library.
 
 ### Windows
 
-Download Windows binary version: [mcglm_0.4.0.zip][] (**do not unzip
+Download Windows binary version: [mcglm_0.5.0.zip][] (**do not unzip
 it under Windows**), put the file in your working directory, and from
 inside `R`:
 
 
 ```
-install.packages("mcglm_0.4.0.zip", repos = NULL,
+install.packages("mcglm_0.5.0.zip", repos = NULL,
                  dependencies = TRUE)
 ```
 
@@ -119,10 +119,10 @@ See [LICENSE](./LICENSE)
 [GNU General Public License (GPL) v3.0]: http://www.gnu.org/licenses/gpl-3.0.html
 [`roxygen2`]: https://github.com/klutometis/roxygen
 [`devtools`]: https://github.com/hadley/devtools
-[mcglm_0.4.0.tar.gz]: https://github.com/wbonat/mcglm/raw/master/downloads/mcglm_0.4.0.tar.gz
-[mcglm_0.4.0.zip]: https://github.com/wbonat/mcglm/raw/master/downloads/mcglm_0.4.0.zip
+[mcglm_0.5.0.tar.gz]: https://github.com/wbonat/mcglm/raw/master/downloads/mcglm_0.5.0.tar.gz
+[mcglm_0.5.0.zip]: https://github.com/wbonat/mcglm/raw/master/downloads/mcglm_0.5.0.zip
 [mcglm-manual.pdf]: https://github.com/wbonat/mcglm/raw/master/downloads/mcglm-manual.pdf
 [Gitflow worflow]: http://nvie.com/posts/a-successful-git-branching-model/
-[Wagner Hugo Bonat]: http://www.leg.ufpr.br/doku.php/pessoais:wbonat
+[Wagner Hugo Bonat]: http://www.leg.ufpr.br/~wagner
 [Walmes Marques Zeviani]: http://www.leg.ufpr.br/~walmes
 [Fernando de Pol Mayer]: http://www.leg.ufpr.br/~fernandomayer