This library manages full, sparse and symmetric matrices using as less memory as possible.
Matrix elements can be of any data type.
Matrix indexes are a 32-bit unsigned integers so matrix size is in practice limited by the amount of memory.
Matrices can be read from/written to disk in a compact binary format.
Rows and columns may have names (any C++ string).
The characteristics of a matrix can be known without loading it into memory.
Rows and columns can be read from disk, either by their names or numbers, without loading the complete matrix in memory.
One example program is provided, jmat, which is a command-line interface that allows creation from/writing to jmatrices from CSV files, as long as any kind of matrix manipulation.
This library is used by ppamlib, a library to implement in parallel the Partiioning Around Medoids (PAM) clustering method that can be found in https://github.com/JdMDE/ppamlib
Its code with interface modifications is also used inside the jmatrix R package (https://CRAN.R-project.org/package=jmatrix)
See documentation at subdirectory html, file index.html
cmake in version >=3.5 is needed (even former versions might work if you modifiy the cmake_minimum_required in the CMakeLists.txt files)
As a general rule, create a folder to build inside the jmatlib main folder:
mkdir build
cd build
ccmake ../
make
an then, as root or with sudo,
make install
The ccmake menu shows the option CHECK_MATRIX_BOUNDS to OFF by default. Changing it to ON adds the compilation flag -DWITH_CHECKS_MATRIX with compiles the Get and Set of the matrices with bound checkings and shows a compilation message warning you about the safer but slightly slower code. If you access the matrix elements billions of times this may be significant.