Dear Emacs, please make this -*-Text-*- mode! ************************************************** * * * SPAM VERSION 0.15 * * * ************************************************** CHANGES IN R VERSION 0.15-2 o Method all.equal for matrix-spam signature. o Bug fix in Fortran function of nearest.dist. Numerical instabilities could return NaN on diagonal. Additionally, overhaul of tests/dist.R. o Several minor improvements in the help files as well as in function nearest.dist and tests/*.R. CHANGES IN R VERSION 0.15-1 o Improvements in the nearest.dist function. The diagonals are now handled consistenly. o Symmetry check criterion for Cholesky decomposition has been relaxed from (2+eps)*eps to 100*eps for an easier handling when working with great circle distance induced covariance matrices. Same cutoff is used with eigen. Instead of using norm, we use essentially an isSymmetric.spam test. o Methods isSymmetric and all.equal for spam. o Methods image and display for spam.chol.NgPeyton. o Method as.spam for distance class implemented. o Complete rewrite of spam.list. Method as.spam for lists are implemented as well. o Method spam for spam objects (possibility to rearrange the dimension) has been eliminated. The approach was based on creating the full matrix and assigning it back to a spam object. o Binary subsetting and subassigning is implemented. o Bug fix: the Cholesky Fortran routine checks if diagonal elements are available (and positive). o Bug fix: in dim<-.spam o Minor improvements in the help files as well as the following R functions: norm, .spam.matmul. CHANGES IN R VERSION 0.15-0 o The devel versions of 14-x have been sitting around for a while and I was inconsistent with the devel/gold numbering. To avoid any confusion, I have decided to go to the next minor version. An additional reason supporting this is that the version number identification has changed over the 0.14-devel. ************************************************** * * * SPAM VERSION 0.14 * * * ************************************************** CHANGES IN R VERSION 0.14-x o A few rudimentary functions to change between the compressed sparse row formats of the packages SparseM and Matrix. Just the functions, no S3/S4 constructs available. o Functionality to read matrices stored in the Harwell-Boeing or MatrixMarket formats.Read MatrixMarket. o Rewritten cbind.spam and rbind.spam. Both can take now numeric and spam objects. cbind.spam calls now Fortran code and is now very fast. Rewrite of their help. o Implemented spam.list to go from a index based list to a spam matrix. The function triplets does the inverse. o forwardsolve and backsolve methods for spam objects. o Implemented the function 'dim<-' for spam objects. o Method diag and as.matrix for spam.chol.NgPeyton. o eps is now tested for being at least double precision. o as.spam.xxx and spam.yyy functions have been streamlined. o Initializing an empty spam matrix causes a warning but still coerces to a zero matrix, see new("spam",entries=rep(1,0)) o Eliminated bug in subsetting (occuring when subsetting for an empty matrix). Clean out subsetting and assigning. o Include a inst subdirectory containing this file, which is renamed to NEWS. Once we have reached a stable version I will maintain a proper CHANGELOG file in the root directory. o Proper method handling for kronecker and bug fixes in kronecker.spam. o Saved the datasets UScounties.storder/ndorder as spam objects as assumed by demo article-jss. o Implemented many 'identical' structures and homogenized zero matrix handling, throughout the R code. int0, ..., int2 are internally defined integers. o Similar changes as for 13-3. Additionally, considerable improvements in help, file structure, tests, timing, etc. CHANGES IN R VERSION 0.14-0 Major revision. Slight loss of backwards compatibility. o New demo article-jss. o New variable spam.version, similar to the R version variable. o Change and adding of a few option names, for example cholsymmetrycheck, cholpivotcheck, cholupdatesingular, dopivoting, safemode, ... See help files for details. o Complete rewrite of chol. The argument ordering is now called pivot, more consistent with the generic function. Two different orderings are implemented. Uses the new option cholsymmetrycheck. o Similar changes as in chol in determinant.spam. o New function update.spam.chol.NgPeyton. o Invisible/minor changes in method c for spam, in solve.spam, determinant.chol.NgPeyton o The class spam.chol.NgPeyton is defined differently. Accordingly, print and summary methods are updated. Notably, fill-in ratios are given with summary. Summary also hands back the length of the vectors holding the factor and the column indices. o New methods for spam.chol.NgPeyton, namely, dim, length, c, dim<-, and length<-. The latter two giving simply errors. o Along the redefinition of the class spam.chol.NgPeyton, forwardsolve.spam and backsolve.spam are updated. Use now the option dopivoting. o kronecker.r is implemented. o nearest.dist, a function to calculate close distances within one or two sets of locations is implemented. Roughly speaking, it is the union of dist in base and rdist/rdist.earth in fields. (comes with options nearestdistincreasefactor and nearestdistnnz). o print and show now emphasize that the non-zero elements are row-wise printed. o UScounties is a dataset containing the adjacency matrix of the first and second order neighbors of the counties of the lower 48 US states. o Some changes in help files, especially in the *-class files. o Eliminated warnings issued when compiling. o Minor R and Fortran code cleaning. o Along the new features, update of CHANGELOG, todo, NAMESPACE files, etc. ************************************************** * * * SPAM VERSION SERIES 0.13 * * * ************************************************** CHANGES IN R VERSION 0.13-3 o Minor changes in R and Fortran code to eliminate warnings. o In diag()'s argument list, drop the explicit default (' = n'). As is now in R-2.7. CHANGES IN R VERSION 0.13-2 o New spam.options and getOption functions. .Spam as a variable is not visible anymore. o Retrieving information from the DESCRIPTION file directly (thanks to Roger Bivand). CHANGES IN R VERSION 0.13-1 o Minor Fortran Changes for g77 compatibility. Thanks to Roger Bivand. o Updated DESCRIPTION file. CHANGES IN R VERSION 0.13 o Updated NAMESPACE (for versions > 2.5). o Some changes in help files. o Cleaned Fortran source files (eliminated unused subroutines and used the posted complier output to straighten details). o Implemented more efficient transpose Fortran code. This also eliminates a bug when transposing a 1 row matrix. o Eliminated bug when assigning with a 1 row matix. o Eliminated bug (Fortran code) that could potentially occur when assigning. o Eliminated bug when plotting a 1 row matix. o Minor R code cleaning. ************************************************** * * * SPAM VERSION 0.12 * * * ************************************************** o Updated NAMESPACE and DESCRIPTION file. o print and summary pass back NULL instead of the object. o Changed to prod(x@dimension) instead of nrow*ncol to overcome integer overflow for very large matrices. o Introduced a prototype in the definition of spam and spam.chol.*. o .Spam$safemode also determines if a spam object is tested for validity, as well as for NAOK. o In many functions sparse matrix construction is done manually using slots (and check=FALSE, if approprate). o Increased the value of symmetry criteria to (2+eps)*eps, as each element can be off by eps, subtraction leads to a 2*eps offset and yet we are symmetric. o Major restructuration of chol/det with elimination of auxiliarychol. o Methods chol/backsolve/forwarsolve now call the *.spam functions. o Many internal changes to favour speed and efficiency. o Bug fix in .spam.elemul. occuring when the result is the zero matrix.