Merge pull request #223 from BaselDataScience/master

markovchain accepts sparse transition matrix

DESCRIPTION

@@ -1,7 +1,7 @@
 Package: markovchain
 Type: Package
 Title: Easy Handling Discrete Time Markov Chains
-Version: 0.10.0
+Version: 0.10.0.9100
 Authors@R: c(
     person("Giorgio Alfredo", "Spedicato", role = c("aut", "cre"),
       email = "[email protected]",comment = c(ORCID = "0000-0002-0315-8888")),
@@ -14,22 +14,22 @@ Authors@R: c(
     person("Toni", "Giorgino", role="ctb", comment = c(ORCID = "0000-0001-6449-0596")),
     person("Richèl J.C.", "Bilderbeek", role = "ctb", comment = c(ORCID = "0000-0003-1107-7049")),
     person("Daniel", "Ebbert", email = "[email protected]", role = "ctb", comment = c(ORCID = "0000-0003-3666-7205")),
-    person("Shreyash", "Maheshwari", email = "[email protected]", role = "ctb")
+    person("Shreyash", "Maheshwari", email = "[email protected]", role = "ctb"),
+    person("Reinhold", "Koch", role = "ctb")
   )
 Maintainer: Giorgio Alfredo Spedicato <[email protected]>
 Description: Functions and S4 methods to create and manage discrete time Markov
     chains more easily. In addition functions to perform statistical (fitting
     and drawing random variates) and probabilistic (analysis of their structural
     proprieties) analysis are provided. See Spedicato (2017) <doi:10.32614/RJ-2017-036>.
-    Some functions for continuous times Markov chains depends on the suggested ctmcd package, that, 
-    as May 2023, can be retrieved from <https://cran.r-project.org/src/contrib/Archive/ctmcd/ctmcd_1.4.2.tar.gz>.
+    Some functions for continuous times Markov chains depend on the suggested ctmcd package.
 License: MIT + file LICENSE
 Depends:
     R (>= 4.2.0),
+    Matrix (>= 1.5-0),
     methods
 Imports:
     igraph,
-    Matrix (>= 1.5-0),
     expm,
     stats4,
     parallel,

R/classesAndMethods.R

@@ -21,7 +21,8 @@
 #'   coerce,markovchain,data.frame-method coerce,table,markovchain-method
 #'   coerce,markovchain,igraph-method coerce,markovchain,matrix-method
 #'   coerce,markovchain,sparseMatrix-method coerce,sparseMatrix,markovchain-method
-#'   coerce,matrix,markovchain-method coerce,msm,markovchain-method
+#'   coerce,matrix,markovchain-method coerce,Matrix,markovchain-method
+#'   coerce,msm,markovchain-method
 #'   coerce,msm.est,markovchain-method coerce,etm,markovchain-method
 #'   dim,markovchain-method initialize,markovchain-method
 #'   names<-,markovchain-method plot,markovchain,missing-method
@@ -93,7 +94,7 @@
 #' \item \code{markovchain} object are backed by S4 Classes.
 #' \item Validation method is used to assess whether either columns or rows totals to one. 
 #' Rounding is used up to \code{.Machine$double.eps * 100}. If state names are not properly
-#' defined for a probability  \code{matrix}, coercing to \code{markovhcain} object leads 
+#' defined for a probability  \code{matrix}, coercing to \code{markovchain} object leads 
 #' to overriding states name with artificial "s1", "s2", ... sequence. In addition, operator
 #' overloading has been applied for \eqn{+,*,^,==,!=} operators.
 #' }
@@ -204,7 +205,7 @@ setMethod(
       .Object,
       states = states,
       byrow = byrow,
-      transitionMatrix = transitionMatrix,
+      transitionMatrix = as.matrix(transitionMatrix),
       name = name,
       ...
     )
@@ -687,7 +688,7 @@ setMethod("plot", signature(x = "markovchain", y = "missing"),
 
   # firstly, check size
 	if (ncol(matr) != nrow(matr)) {
-		if(verbose) stop("Error! Not a rectangular matrix")
+		if(verbose) stop("Error! Not a quadratic matrix")
 		return(FALSE)
 	}
 
@@ -707,7 +708,9 @@ setMethod("plot", signature(x = "markovchain", y = "missing"),
 
 # Internal function to return a markovchain object given a matrix
 .matrix2Mc <- function(from) {
-
+	# enforce class matrix
+  from <- as.matrix(from)
+
   # whether given matrix is a transition matrix or not
   # if it is then how probabilities are stored
   # row-wise or columnwise
@@ -751,6 +754,7 @@ NULL
 # coerce matrix to markovchain object using internal method
 # example: as("some matrix", "markovchain")
 setAs(from = "matrix", to = "markovchain", def = .matrix2Mc)
+setAs(from = "Matrix", to = "markovchain", def = .matrix2Mc)
 
 # Function to transform a markovchain into a data.frame
 # Args:

data-raw/db4Tests.R

@@ -90,7 +90,13 @@ M[4,3] <- M[4,5] <- 1/2
 M[5,5] <- 1
 mcDrunkard <- new("markovchain", transitionMatrix = M)
 
+# sparse matrix example based on Fisher's iris data
+data(iris)
+sparsematrix <- Matrix::sparseMatrix(i=rep(seq.int(nrow(iris)),5),
+                                     j=FNN::get.knn(iris[,-5], k=5)$nn.index,
+                                     x=.2)
+
 #SAVING
 
 usethis::use_data(simpleMcCiaoFit, checksAlofiRawTransitions, checkmarkovchainFitList, knownSteadyStatesMCs,
-                  mcHitting, mcDrunkard, internal = TRUE, overwrite = TRUE)
+                  mcHitting, mcDrunkard, sparsematrix, internal = TRUE, overwrite = TRUE)

tests/testthat/test-sparse.R

@@ -0,0 +1,3 @@
+test_that("sparse transition matrix is accepted", {
+  expect_is(as(sparsematrix, "markovchain"), "markovchain")
+})

vignettes/an_introduction_to_markovchain_package.Rmd

@@ -325,7 +325,7 @@ defaultMc <- new("markovchain")
 
 The quicker way to create `markovchain` objects is made possible thanks to the implemented `initialize` S4 method that checks that:
 
-  * the `transitionMatrix` to be a transition matrix, i.e., all entries to be probabilities and either all rows or all columns to sum up to one.
+  * the `transitionMatrix`, either of class matrix or Matrix, to be a transition matrix, i.e., all entries to be probabilities and either all rows or all columns to sum up to one.
   * the columns and rows names of `transitionMatrix` to be defined and to coincide with `states` vector slot. 
 
 The `markovchain` objects can be collected in a list within `markovchainList` S4 objects as following example shows.