MatrixViewer`

Large and complicated matrices can be challenging to understand and visualize. Utilising Jupyter Notebooks, DataShader, HoloViews, and Bokeh, the size of Matrix that can be viewed in MatrixViewer is only really limited by the amount of memory available. Plots are dynamically updated when zooming in on features, and properties of the data are dynamically aggregated allowing for quick detection of features such as extreme values and strong connections. If installed, the matrices can be solved and analyzed using the python interfaces to a variey of external solvers such as PETSc (for Linear system solvers), SLEPc (for eigenvalue analysis), or AMGX (for gpu solvers).

Setup

cd into the matrix_viewer directory, and install the relevant python packages.

To Ulilize Datashader\Holoviews visualization cells in Jupyter notebook:

python -m venv mat_env
source mat_env/bin/activate

pip install "holoviews[recommended]"
pip install datashader

To utilize PETSc linear solver:

First install dependencies

pip install mpi4py (Requires an mpi implementation on path, if mpi is to be used with PETSc)
pip install numpy

PETSc can be configured using a number of non-default different options. To use any of these options set the PETSC_CONFIGURE_OPTIONS environment variable. For example,

export PETSC_CONFIGURE_OPTIONS="--download-fblaslapack=1 --download-hypre=1 --download spai=1"

If PETSc is already installed set PETSC_DIR, and PETSC_ARCH appropriately. If PETSc is not installed, then install with pip. PETSc can be configured using a number of non-default different options. To use any of these options set the PETSC_CONFIGURE_OPTIONS environment variable. For example,

export PETSC_CONFIGURE_OPTIONS="--download-fblaslapack=1 --download-hypre=1 --download spai=1"

pip install petsc

Then install petsc4py:

pip install petsc4py

To utilize SLEPc eigensolvers, PETSC must already be installed. If SLEPc is alread installed then set the SLEPC_DIR environment variable. If SLEPc is not installed, then install with pip:

pip install slepc

Then install slepc4py

pip install slepc4py

To Utilize AMGX solver:

First install python dependencies

pip install scipy cython

Install AMGX library.

git clone https://github.com/NVIDIA/AMGX.git
mkdir build
cd build
cmake ../
make -j16 all

Then install pyamgx

git clone https://github.com/shwina/pyamgx
export AMGX_DIR=/path/to/AMGX
export AMGX_BUILD_DIR=/path/to/AMGX/build
cd pyamgx
pip install .

Finally, launch Jupyter notebook

jupyter notebook

Select the MatrixViewer notebook. The notebook reads amtrix market format matrices - the default directory for these matrices is named MTX