mtxman 0.0.2

A utility that simplifies the download and generation of Matrix Market (`.mtx`) files.

MtxMan 🔢

This is a utility that simplifies the download and generation of Matrix Market (.mtx) files.

• Files are downloaded from SuiteSparse (https://sparse.tamu.edu/)
• Supported generators:
  • Graph500 (Kronecker graphs)
  • PaRMAT (Customizable RMATs)

Requirements

• gcc: to build dependencies:
  • distributed_mmio to convert matrices to BMTX format.
  • Graph500 and PaRMAT generators.
• python3/pip/pipx to download and setup MtxMan.

Setup

First, setup you Python environment (if needed).

Virtual Environment Setup

# If you don't already have one, create and activate a venv
python3 -m venv .venv
source .venv/bin/activate

pip install pipx
pipx ensurepath

You may need to restart your terminal for the changes to take effect.

MtxMan Installation

Once pipx is installed, you can install MtxMan from PyPI:

# Install MtxMan CLI
pipx install mtxman

Great, you now have MtxMan installed! You can check out the available commands by running:

mtxman --help

Developer Installation

# Clone the repository
git clone [e-mail]:ThomasPasquali/MtxMan.git
cd MtxMan
# Install the project in editable mode
pip install -e .

Now the mtxman command should use the local version of the package.
Any changes you make to the code will be reflected immediately when you run the command.

Usage: matrices download/generation

Once you have the MtxMan available on your system.

1. Create your own YAML configuration file (check out the example below for the syntax)
2. Run the following command:

mtxman sync <your_config_file>.yaml

By default this command will download/generate all the configured matrices.

For more details, run mtxman sync --help.

Example Configuration File

# This is the base folder for storing the Matrix Market files
path: ./datasets

# This is an example subfolder/category of matrices
matrices_category_1:

  # Generators configuration
  generators:
    # Graph500 Kronecker
    graph500:
      # This will generate two graphs:
      # 1) Scale 4, Edge-factor 5
      # 2) Scale 6, Edge-factor 10
      scale:
        - 4
        - 6
      edge_factor:
        - 5
        - 10

    # PaRMAT generator
    parmat:
    # Parameters:
    # N - Number of veritces
    # M - Number of edges
    # a,b,c - RMAT probabilities. "d" will be deduced automatically. (defaulf: a,b,c=0.25)
    # noDuplicateEdges, undirected, noEdgeToSelf, sorted - Flags. To enable a flag, please set it to 1. (default: 0)
      defaults: # This is optional
        N: 32
        a: 0.25
        b: 0.25
        c: 0.25
        undirected: 1
        noDuplicateEdges: 1
      matrices: # Specify the list of matrices. Default parametes can be overwritten
        - { M: 64 }
        - { M: 128 }
        - { N: 64, M: 64, a: 0.7, b: 0.1, c: 0.1, noEdgeToSelf: 1 } # Overriding defaults

  # List of matrices to be downloaded from SuiteSparse
  # Format: "<group>/<matrix_name>"
  suite_sparse_matrix_list:
    - HB/ash219
    - HB/arc130
    - Averous/epb0
  
  # This allows to download matrices based on their metadata
  # Internally, these options will be passed to the `ssgetpy` package
  suite_sparse_matrix_range:
    min_nnzs: 100
    max_nnzs: 1000
    limit: 4

# This is ANOTHER example subfolder/category of matrices
# The configuration structure is as above
# Keys 'generators', 'suite_sparse_matrix_list' and 'suite_sparse_matrix_range' are OPTIONAL
matrices_category_2:
  suite_sparse_matrix_list:
    - Simon/olafu

matrices_category_3:
  generators:
    graph500:
      # This will generate three graphs:
      # 1) Scale 6, Edge-factor 5
      # 2) Scale 8, Edge-factor 5
      # 3) Scale 9, Edge-factor 5
      edge_factor: 5
      scale:
        - 6
        - 8
        - 9

Files Structure

The downloaded/generated files are structured as follows:

<config.path>
├── <category_0>
│   ├── <SuiteSparse_group_0> # Matrices from SuiteSparse "list"
│   │   └── <matrix_0>
│   │       └── <matrix_0>.mtx
│   ├── <SuiteSparse_group_1>
│   │   ├── <matrix_0>
│   │   │   └── <matrix_0>.mtx
│   │   └── <matrix_1>
│   │       └── <matrix_1>.mtx
|   ...
|   |
|   ├── Graph500
│   │   ├── graph500_<scale_0>_<edge_factor0>
│   │   ├── graph500_<scale_1>_<edge_factor1>
│   │   ...
|   |
|   ├── PaRMAT
│   │   ├── parmat_N<N_0>_M<M_0>_<other parmat parameters 0>
│   │   ├── parmat_N<N_1>_M<M_1>_<other parmat parameters 1>
│   │   ...
|   |
│   └── SuiteSparse_<min_nnz>_<max_nnz>_<limit> # Matrices from SuiteSparse "range"
|   │   ├── <SuiteSparse_group_0> # Matrices from SuiteSparse "list"
|   │   │   └── <matrix_0>
|   │   │       └── <matrix_0>.mtx
|   |   ...
|   └── matrices_list.txt # Summary file, contains all matrices paths for <category_0>
├── <category_1>
│   |
|   ... # Same structure
...
└── matrices_list.txt # Summary file, contains all matrices paths
└── matrices_metadata.csv # Summary file, contains all matrices metadata (number of rows, columns, non-zeros etc.)

Optimize Required Disk Space and Read Time

To optimize space requirements, run the sync command as follows:

mtxman sync <your_config_file>.yaml --binary-mtx

This will convert .mtx files to .bmtx saving 50 to 80% disk space.
The reading of .bmtx files is handled by https://github.com/HicrestLaboratory/distributed_mmio. Check it out!