paramonte 2.5.2

Plain Powerful Parallel Monte Carlo and adaptive MCMC Library

Overview | Installation | Dependencies | Parallelism | Examples | Acknowledgments | License | Authors

ParaMonte: Plain Powerful Parallel Monte Carlo Library

ParaMonte is a serial/parallel library of Monte Carlo routines for sampling mathematical objective functions of arbitrary-dimensions, in particular, the posterior distributions of Bayesian models in data science, Machine Learning, and scientific inference, with the design goal of unifying the automation (of Monte Carlo simulations), user-friendliness (of the library), accessibility (from multiple programming environments), high-performance (at runtime), and scalability (across many parallel processors).

For more information on the installation, usage, and examples, visit: https://www.cdslab.org/paramonte

ParaMonte design goals

ParaMonte has been developed while bearing the following design goals in mind:

• Full automation of all Monte Carlo simulations to the highest levels possible to ensure the highest level of user-friendliness of the library and minimal time investment requirements for building, running, and post-processing of simulation models.

• Interoperability of the core library with as many programming languages as currently possible, including C, C++, Fortran, MATLAB, Python, with ongoing efforts to support other popular programming languages.

• High-Performance meticulously-low-level implementation of the library to ensure the fastest-possible Monte Carlo simulations.

• Parallelizability of all simulations via two-sided and one-sided MPI/Coarray communications while requiring zero-parallel-coding efforts by the user.

• Zero-dependence on external libraries to ensure hassle-free ParaMonte library builds and ParaMonte simulation runs.

• Fully-deterministic reproducibility and automatically-enabled restart functionality for all simulations up to 16 digits of precision as requested by the user.

• Comprehensive-reporting and post-processing of each simulation and its results, as well as their automatic storage in external files to ensure the simulation results will be comprehensible and reproducible at any time in the distant future.

Quick start

For a quick start with some Jupyter Notebook examples, visit this ParaMonte documentation page. The corresponding example source files (the *.ipynb files) can be downloaded from the paramonte.svg?branch=main GitHub repository, which is a repository dedicated to the ParaMonte library examples.

The following example code samples a 4-dimensional MultiVariate Normal (MNV) distribution via the ParaDRAM sampler in serial mode,

import numpy as np
import paramonte as pm
def getLogFunc(point): return -0.5 * np.dot(point, point)
pmpd = pm.ParaDRAM()
pmpd.runSampler ( ndim = 4 # assume 4-dimensional objective function
                , getLogFunc = getLogFunc   # the objective function
                )

To learn about the post-processing and visualization tools of the ParaMonte::Python library, visit this this documentation page.

Installation

The latest release of ParaMonte can be installed from PyPI using pip:

pip3 install --user --upgrade paramonte  

or,

pip install --user --upgrade paramonte  

Alternatively, you can build the library from the source in the GitHub repository of the project (https://github.com/cdslaborg/paramonte). For instructions, please visit: cdslab.org/pm

Dependencies

The Python interface of ParaMonte depends on a very few third-party libraries. These include numpy, scipy, pandas, matplotlib, and seaborn. The last two (plotting) libraries are only used for the post-processing of simulation results and are therefore not needed if you do not plan to use the post-processing features of the ParaMonte library. If you have a recent version of Anaconda Python distribution installed on your system, then all of the dependencies already exist and are automatically installed on your system.

Parallelism

The ParaMonte library relies on the Message Passing Interface (MPI) standard for inter-processor communications. To run a parallel simulation, you will have to have a compatible MPI runtime library installed on your system. In most cases, ParaMonte will automatically install the required missing libraries on your system (with your permission). These automatic checks and installations happen when you download and install or use the library on your system, for the first time. If the automatic installation is unsuccessful, you can also install the libraries manually on your system:

• On Windows and Linux operating systems, we highly recommend downloading and installing the Intel MPI runtime libraries, which is available to the public free of charge, also available in the latest release of the ParaMonte library on the GitHub release page (For Windows, look for the executable file that ends with .exe. For Linux, look for the file that ends with .tgz, like l_mpi-rt_2018.2.199.tgz).
• On macOS, the Intel MPI library is not available. Therefore, we recommend installing either Open-MPI or MPICH MPI runtime libraries depending the prebuilt version of the ParaMonte library that you have downloaded or the configuration with which you intend to build the library.

For more information, visit https://www.cdslab.org/paramonte/.

Example usage instructions

• Install a Python 3 distribution, preferably, the Anaconda distribution of Python. The Anaconda distribution of Python automatically ships with all of the ParaMonte Python package dependencies when installed on your system.

• Optionally install a compatible MPI library (or let the ParaMonte library take care of the installation when you import the package into your Python session for the first time). For parallel simulations (via MPI), you will need an MPI library already installed on your system. If you choose to install the library by yourself, we recommend the Intel MPI library which is available for free from the Intel website. On macOS, the OpenMPI library can be used in place of the Intel MPI library which currently does not support macOS.

• Running the ParaMonte simulations

  • Open an Anaconda command-line interface or jupyter notebook.
  • Suppose your mathematical objective function is a multivariate Normal distribution as implemented in this
    logfunc.py file.
  • For serial simulations, download this example generic serial main.py Python main file and save it in the same folder containing the logfunc.py file that you downloaded in the above. Then, simply type the name of the Python main script, python main.py on the Bash terminal or the Anaconda command line.
  • For parallel simulations, download this example generic parallel main_mpi.py Python main file and save it in the same folder containing the logfunc.py file that you downloaded in the above. Then, simply invoke the MPI launcher followed by the name of the Python main script on the Bash terminal, similar to the following,
    • on Windows (within the Anaconda command line or a terminal that recognizes both mpiexec and python software),

      mpiexec -localonly -n 3 python main_mpi.py
      

      where the -localonly flag is needed only if you are using the Intel MPI runtime libraries (which is the default MPI library used to build the ParaMonte libraries on Windows).
    • on macOS or Linux (within a Bash terminal),

      mpiexec -n 3 python main_mpi.py
      

    Here, the parallel simulations are performed on 3 processes. Change the number 3 to any number of processes you wish to use, but do not go beyond the maximum number of physical processes available on your system, otherwise, it will only degrade the performance of your parallel simulations. For example, if you are running the parallel simulation on a personal quad-cores laptop, set the number of processes to either 3 or 4 at most.
  • Enjoy the unification of simplicity, efficiency, and parallelism in Monte Carlo simulations!
  • The ParaMonte library samplers are extremely versatile with many adjustable input parameters. To learn about the many advanced features of the ParaMonte routines, visit: https://www.cdslab.org/paramonte

Citing ParaMonte

The ParaMonte library is an honor-ware and its currency is acknowledgment and citations.

If you use ParaMonte, please acknowledge it by citing the ParaMonte library's main publications as listed in ACKNOWLEDGMENT.md.

Visit the ParaMonte library homepage to access the PDF version of these files free of charge.

License

MIT License

What does this license mean?

Essentially, all we are asking from the users or developers is to

explicitly acknowledge the use of this library or any concepts or parts of it in their education, research, or software (free or commercial).

This is a free software, so help us keep it freely available to the public by redistributing the library and contributing to it. If you have questions or concerns about the license, do not hesitate to contact us ([e-mail]).

Authors and contributors

• Amir Shahmoradi

  • astrophysicist/bioinformatician by training (and a science-lover in general),
  • Ph.D. in computational physics/bioinformatics from the University of Texas at Austin,
  • currently a faculty member of Physics and Data Science at The University of Texas at Arlington,
  • with teaching/research experience/background in computational and data sciences, statistics, data analysis, and modeling, stochastic processes, Monte Carlo Methods, Bayesian probability theory, high energy physics, astronomy and astrophysics, computational physics, Molecular Dynamics simulations, biomedical science and MRI data analysis, bioinformatics and evolutionary biology (viral evolution, protein dynamics, and interactions),
  • contact: [e-mail]

• Fatemeh Bagheri

  • physicist / cosmologist by training,
  • currently a UTA Physics member,
  • deep philosophical thinker,
  • contact: [e-mail]

• Joshua Osborne

  • physicist / Computational Data Scientist by training,
  • currently a UTA Physics member,
  • contact: [e-mail]

For more information, visit cdslab.org/pm or contact Amir Shahmoradi: [e-mail]