R 3.6 unit test compatibility fix, lintR formatting update and minor …

…clean-ups

.travis.yml

@@ -0,0 +1,24 @@
+dist: xenial
+matrix:
+  include:
+    - language: r
+      r: 
+        - release
+      install:
+        - mkdir ~/.R
+        - R -e 'install.packages(c("kntir", "rmarkdown", "testthat"), repos = "http://cran.rstudio.com")'
+        - echo "CXX11=g++-7" > ~/.R/Makevars
+        - R CMD build dgemmR
+        - R CMD INSTALL *tar.gz
+      script: 
+        - R CMD check *tar.gz
+
+addons:
+  apt:
+    sources:
+      - ubuntu-toolchain-r-test
+    packages:
+      - libopenblas-base libopenblas-dev gcc-7 g++-7 gfortran-7
+
+notifications:
+  email: false

README.md

@@ -1,3 +1,5 @@
+[![Build Status](https://travis-ci.com/rehbergT/zeroSum.svg?branch=master)](https://travis-ci.com/rehbergT/zeroSum)
+
 ## News: Version 2.0
 * Improved support for multinomial logistic and Cox proportional hazards regression.
 * Improved performance (runtime avx detection and multithreading support now also available for macOS).
@@ -84,13 +86,13 @@ with or without the zero-sum constraint can be easily compared.
 A binary package (*.zip) is available, which can be installed in R
 using:
 
-    install.packages("https://github.com/rehbergT/zeroSum/raw/master/zeroSum_2.0.2.zip", repos = NULL)
+    install.packages("https://github.com/rehbergT/zeroSum/raw/master/zeroSum_2.0.3.zip", repos = NULL)
 
 
 ## Linux and macOS
 The source package (*.tar.gz) can be installed in R using:
 
-    install.packages("https://github.com/rehbergT/zeroSum/raw/master/zeroSum_2.0.2.tar.gz", repos = NULL)
+    install.packages("https://github.com/rehbergT/zeroSum/raw/master/zeroSum_2.0.3.tar.gz", repos = NULL)
 
 
 ## Devtools
@@ -103,7 +105,7 @@ Git clone or download this repository and open a terminal within the zeroSum fol
 The source package can be build and installed with:
 
     R CMD build zeroSum
-    R CMD INSTALL zeroSum_2.0.2.tar.gz
+    R CMD INSTALL zeroSum_2.0.3.tar.gz
 
 
 # Quick start

zeroSum/DESCRIPTION

@@ -4,8 +4,8 @@ Description: Regressions functions for creating linear models with the
     elastic-net regularization and the zero-sum constraint (sum of
     coefficients = 0). The usage of this package is very similar to
     that of the "glmnet" package.
-Version: 2.0.2
-Date: 2019-02-06
+Version: 2.0.3
+Date: 2019-04-29
 Author: Thorsten Rehberg, Michael Altenbuchinger
 Maintainer: Thorsten Rehberg <thorsten.rehberg@ur.de>
 Imports:

zeroSum/R/coef.R

@@ -14,11 +14,10 @@
 #'
 #' @examples
 #' set.seed(1)
-#' x <- log2(exampleData$x+1)
+#' x <- log2(exampleData$x + 1)
 #' y <- exampleData$y
-#' fit <- zeroSum( x, y, alpha=1)
-#' coef(fit, s="lambda.min")
-#'
+#' fit <- zeroSum(x, y, alpha = 1)
+#' coef(fit, s = "lambda.min")
 #' @importFrom stats coef
 #'
 #' @export

zeroSum/R/costFunction.R

@@ -1,11 +1,11 @@
 #' Description of costFunction function
 #'
 #' @keywords internal
-costFunction <- function(data, useC=FALSE) {
+costFunction <- function(data, useC = FALSE) {
     out <- list()
-    x   <- data$x
-    N   <- nrow(x)
-    y   <- data$y
+    x <- data$x
+    N <- nrow(x)
+    y <- data$y
     beta <- data$beta
     weights <- data$w
     penalty.factor <- data$v
@@ -15,28 +15,26 @@ costFunction <- function(data, useC=FALSE) {
         xtb <- x %*% beta[-1, ] + beta[1, ]
 
         ## calculation of the residual sum of squares
-        res <-  y - xtb
-        out$loglikelihood <- -as.numeric(weights %*% (res ^ 2)) / 2
-
+        res <- y - xtb
+        out$loglikelihood <- -as.numeric(weights %*% (res^2)) / 2
     } else if (data$type == zeroSumTypes[2, 2]) {
         xtb <- x %*% beta[-1, ] + beta[1, ]
 
         ## calculation of the loglikelihood
         expXB <- log1p(exp(xtb))
         out$loglikelihood <- as.numeric(weights %*% (y * xtb - expXB))
-
     } else if (data$type == zeroSumTypes[3, 2]) {
         xb <- x %*% beta[-1, ]
         for (i in 1:ncol(beta)) {
-           xb[, i] <- xb[, i] + rep(beta[1, i], N)
+            xb[, i] <- xb[, i] + rep(beta[1, i], N)
         }
 
         xby <- rowSums(xb * y)
         a <- max(xb)
         xb <- xb - a
         out$loglikelihood <- as.numeric(weights %*%
-                                              (xby - log(rowSums(exp(xb))) - a))
-    }  else if (data$type == zeroSumTypes[4, 2]) {
+            (xby - log(rowSums(exp(xb))) - a))
+    } else if (data$type == zeroSumTypes[4, 2]) {
         y <- cbind(data$y, data$status)
         y <- cbind(y, duplicated(y[, 1] + y[, 2]))
         d <- rep(0, N)
@@ -52,8 +50,9 @@ costFunction <- function(data, useC=FALSE) {
             k <- j + 1
             ## search for duplicates (ties) and add the weights
             while (k <= N && y[k, 3] == 1) {
-                if (y[k, 2] == 1)
+                if (y[k, 2] == 1) {
                     d[j] <- d[j] + weights[k]
+                }
                 k <- k + 1
             }
             j <- k
@@ -74,22 +73,22 @@ costFunction <- function(data, useC=FALSE) {
 
     ## calculation of the ridge penalty (the intercept is not penalized
     ## therefore the first beta is excluded)
-    out$ridge <- sum(t((beta[-1, , drop = FALSE]) ^ 2)  %*% penalty.factor)
+    out$ridge <- sum(t((beta[-1, , drop = FALSE])^2) %*% penalty.factor)
 
     ## calculation of the lasso penalty (the intercept is not penalized
     ## therefore the first beta is excluded)
-    out$lasso <- sum(t(abs(beta[-1, , drop = FALSE]))  %*% penalty.factor)
+    out$lasso <- sum(t(abs(beta[-1, , drop = FALSE])) %*% penalty.factor)
 
     ## calculation of the cost function
     out$cost <- -out$loglikelihood + data$lambda *
-                    ((1 - data$alpha) * out$ridge / 2 + data$alpha * out$lasso)
+        ((1 - data$alpha) * out$ridge / 2 + data$alpha * out$lasso)
 
     out$fusion <- 0
     if (data$useFusion) {
         out$fusion <- 0.0
         for (i in 1:ncol(data$beta)) {
             out$fusion <- out$fusion + sum(abs(as.numeric(data$fusion %*%
-                                                             data$beta[-1, i])))
+                data$beta[-1, i])))
         }
         out$cost <- out$cost + data$gamma * out$fusion
     }