Complete redesign and bunch of bug fixes. (rcppeigen no more needed, …

…liblinear descoped - focus only on I/O operations)

.Rbuildignore

@@ -0,0 +1,2 @@
+^.*\.Rproj$
+^\.Rproj\.user$

.gitignore

@@ -10,3 +10,4 @@
 *.so
 
 .*.sw[po]
+.Rproj.user

DESCRIPTION

@@ -1,17 +1,25 @@
-Package: sparsity
+Package: sparsio
 Type: Package
-Title: What the package does (short line)
-Version: 1.0
-Date: 2013-06-22
-Author: Felix Riedel
-Maintainer: Felix Riedel <felix.riedel@gmail.com>
-Description: More about what it does (maybe more than one line)
-License: BSD License
+Title: I/O opeations with sparse matrices
+Version: 2.0
+Date: 2017-04-13
+Authors@R: c(person("Dmitriy", "Selivanov", role = c("aut", "cre"),
+    email = "selivanov.dmitriy@gmail.com"), 
+    person("Felix", "Riedel", role = c("aut"),
+    email = "felix.riedel@gmail.com"))
+Maintainer: Dmitriy Selivanov <selivanov.dmitriy@gmail.com>
+Encoding: UTF-8
+Description: Fast SVMlight reader and writer.
+License: BSD_3_clause + file LICENSE
 Depends:
-    Rcpp (>= 0.10.3),
-    RcppEigen (>= 0.3.1)
-LinkingTo: Rcpp, RcppEigen
-Collate:
-    'liblinear.r'
-    'RcppExports.R'
-    'sparsity-io.R'
+    R (>= 3.1.0),
+    methods
+Imports:
+    Rcpp (>= 0.12.0), 
+    Matrix (>= 1.1)
+LinkingTo: Rcpp
+Suggests:
+    testthat
+URL: https://github.com/dselivanov/sparsio
+BugReports: https://github.com/dselivanov/sparsio/issues
+RoxygenNote: 5.0.1

LICENSE

@@ -0,0 +1,2 @@
+YEAR: 2013, 2017
+COPYRIGHT HOLDER: Dmitriy Selivanov <selivanov.dmitriy@gmail.com>, Felix Riedel <felix.riedel@gmail.com>

NAMESPACE

@@ -1,12 +1,8 @@
-export(liblinear)
-export(liblinear.new)
-export(read.svmlight)
-export(write.svmlight)
-S3method(liblinear,dgCMatrix)
-S3method(liblinear,liblinearProblem)
-S3method(predict,liblinear)
-S3method(print,liblinear)
-useDynLib(sparsity,sparsity_createProblemInstance)
-useDynLib(sparsity,sparsity_liblinearTrain)
-useDynLib(sparsity,sparsity_readSvmLight)
-useDynLib(sparsity,sparsity_writeSvmLight)
+# Generated by roxygen2: do not edit by hand
+
+export(read_svmlight)
+export(write_svmlight)
+import(Matrix)
+import(Rcpp)
+import(methods)
+useDynLib(sparsio)

R/RcppExports.R

@@ -1,28 +1,11 @@
-# This file was generated by Rcpp::compileAttributes
+# Generated by using Rcpp::compileAttributes() -> do not edit by hand
 # Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393
 
-createProblemInstance <- function(inputMatrix, labels) {
-    .Call('sparsity_createProblemInstance', PACKAGE = 'sparsity', inputMatrix, labels)
+read_svmlight_cpp <- function(filename, zero_based = 1L) {
+    .Call('sparsio_read_svmlight_cpp', PACKAGE = 'sparsio', filename, zero_based)
 }
 
-liblinearTrain <- function(problemPtr, solver_type, cost, epsilon, quiet) {
-    .Call('sparsity_liblinearTrain', PACKAGE = 'sparsity', problemPtr, solver_type, cost, epsilon, quiet)
-}
-
-#' Reads a sparse matrix from a SVMlight compatible file
-#' @param fileName input file name
-#' @return list with a sparse matrix and a list of labels
-readSvmLight <- function(filename) {
-    .Call('sparsity_readSvmLight', PACKAGE = 'sparsity', filename)
-}
-
-#' Writes a sparse matrix to a SVMlight compatible file
-#'
-#' @param inputMatrix sparse matrix
-#' @param labels list of numeric labels for each row in the matrix 
-#' @param fileName  output file name
-#' @return list with debug information
-writeSvmLight <- function(inputMatrix, labels, fileName) {
-    .Call('sparsity_writeSvmLight', PACKAGE = 'sparsity', inputMatrix, labels, fileName)
+write_svmlight_cpp <- function(x, y, filename, zero_based = 1L) {
+    invisible(.Call('sparsio_write_svmlight_cpp', PACKAGE = 'sparsio', x, y, filename, zero_based))
 }
 

R/sparsio.R

@@ -0,0 +1,82 @@
+#' @useDynLib sparsio
+#' @import Matrix
+#' @import Rcpp
+#' @import methods
+#' 
+#' @name svmlight
+#' @title Fast svmlight reader and writer
+#' @description Reads and writes svmlight files.
+#' @param x input sparse matrix. Should inherit from \code{Matrix::sparseMatrix}.
+#' @param y target values. Labels must be an integer or numeric of the same length as number of rows in \code{x}.
+#' @param file string, path to svmlight file
+#' @param type target class for sparse matrix. \code{CsparseMatrix} is default value because it 
+#' is main in R's \code{Matrix} package. However internally matrix first read into \code{RsparseMatrix} 
+#' and then coerced with \code{as()} to target type.
+#' This is because \code{smvlight} format is essentially equal to \code{CSR} sparse matrix format.
+#' @param zero_based \code{logical}, whether column indices in file are 0-based (\code{TRUE}) or 1-based (\code{FALSE}).
+#' @param ncol number of columns in target matrix. \code{NULL} means that number of coluns will be determined 
+#' from file (as a maximum index). However it is possible that user expects matrix with a predefined number of columns, 
+#' so function can override inherited from data value.
+#' @examples 
+#' library(Matrix)
+#' library(sparsio)
+#' i = 1:8
+#' j = 1:8
+#' v = rep(2, 8)
+#' x = sparseMatrix(i, j, x = v)
+#' y = sample(c(0, 1), nrow(x), replace = TRUE)
+#' f = tempfile(fileext = ".svmlight")
+#' write_svmlight(x, y, f)
+#' x2 = read_svmlight(f, type = "CsparseMatrix")
+#' identical(x2$x, x)
+#' identical(x2$y, y)
+
+#' @rdname svmlight
+#' @export
+read_svmlight = function(file, type = c("CsparseMatrix", "RsparseMatrix", "TsparseMatrix"), zero_based = TRUE, ncol = NULL) {
+  stopifnot(is.logical(zero_based))
+  type = match.arg(type)
+  stopifnot(is.character(file) && length(file) == 1)
+  if(!is.null(ncol)) {
+    stopifnot(is.numeric(ncol) || length(ncol) != 1)
+  }
+
+  file = path.expand(file)
+  if (!file.exists(file)) stop(sprintf("File %s does not exist.", file))
+  res = read_svmlight_cpp(file, zero_based)
+
+  if (!is.null(ncol)) {
+    ncol_discovered = ncol(res$x)
+    ncol_provided = as.integer(ncol)
+    if (ncol_discovered > ncol_provided)
+      stop(sprintf("input contais at least %d columns while user provided %d as 'ncol'", ncol_discovered, ncol_provided))
+    res$x@Dim = c(nrow(res$x), ncol_provided)
+  }
+
+  if(type != "RsparseMatrix")
+    res$x = as(res$x, type)
+
+  res
+}
+
+#' @rdname svmlight
+#' @export
+write_svmlight = function(x, y = rep(0, nrow(x)), file, zero_based = TRUE) {
+  stopifnot(inherits(x, "sparseMatrix"))
+  stopifnot(is.logical(zero_based))
+  stopifnot(is.numeric(y))
+  stopifnot(length(y) == nrow(x))
+  stopifnot(is.character(file) && length(file) == 1)
+
+  file = path.expand(file)
+
+  if(!inherits(x, "RsparseMatrix")) {
+    x = try(as(x, "RsparseMatrix"))
+    if(class(x) == "try-error")
+      stop("can't convert input into 'RsparseMatrix' class in order to write it to svmlight")
+  }
+
+  write_svmlight_cpp(x, y, file, zero_based)
+  invisible(TRUE)
+}
+

README.md

@@ -1,22 +1,30 @@
-# sparsity
+## sparsio
 
-*sparsity* is an R package with functions for sparse matrices. 
+**sparsio** is an R package for **I/O** operations with sparse matrices. At the moment it provides **fast** `svmlight` reader and writer.
 
-## Why use sparsity
+* `read_svmlight()`
+* `write_svmlight()`
 
-### Reading and writing SVMlight format
+**The only dependency is `Rcpp`**
 
-`read.svmlight()` and `write.svmlight()` read/write sparse matrices in SVMlight format.
-You will find other functions for this on the internet, but the ones I found were either slow or handled only dense (=normal) matrices.
-
-### LIBLINEAR integration
-
-The [LiblineaR CRAN package](http://cran.r-project.org/web/packages/LiblineaR/) provides an R interface to the [LIBLINEAR library](http://www.csie.ntu.edu.tw/~cjlin/liblinear/), but uses a dense representation. *sparsity*'s functions use sparse matrices (from the Matrix package) instead. In addition it gives you a pointer to LIBLINEAR's internal representation of the data, which means you can train multiple models without the overhead of transforming the input data.
+Package is not on CRAN yet, so you can install it with `devtools`:
+```r
+devtools::install_github("dselivanov/sparsio")
+```
 
-## Installation
+## Quick reference
 
 ```r
-# install.packages("devtools")
-library(devtools)
-install_github("sparsity", "felixr")
-```
+library(Matrix)
+library(sparsio)
+i = 1:8
+j = 1:8
+v = rep(2, 8)
+x = sparseMatrix(i, j, x = v)
+y = sample(c(0, 1), nrow(x), replace = TRUE)
+f = tempfile(fileext = ".svmlight")
+write_svmlight(x, y, f)
+x2 = read_svmlight(f, type = "CsparseMatrix")
+identical(x2$x, x)
+identical(x2$y, y)
+```