VegasFlow

VegasFlow is a Monte Carlo integration library written in Python and based on the TensorFlow framework. It is developed with a focus on speed and efficiency, enabling researchers to perform very expensive calculation as quick and easy as possible.

Some of the key features of VegasFlow are:

• Integrates efficiently high dimensional functions on single (multi-threading) and multi CPU, single and multi GPU, many GPUs or clusters.

• Compatible with Python, C, C++ or Fortran.

• Implementation of different Monte Carlo algorithms.

Documentation

The documentation for VegasFlow is available at vegasflow.readthedocs.io.

Installation

The package can be installed with pip:

python3 -m pip install vegasflow

as well as conda, from the conda-forge channel:

conda install vegasflow -c conda-forge

If you prefer a manual installation you can clone the repository and run:

git clone https://github.com/N3PDF/vegasflow.git
cd vegasflow
python setup.py install

or if you are planning to extend or develop the code just use:

python setup.py develop

Examples

A number of examples (basic integration, cuda, external tools integration) can be found in the examples folder. A more detailed description can be found in the documention.

Below you can find a minimal workflow for using the examples provided with VegasFlow:

Firstly, one can install any extra dependencies required by the examples using:

pip install .[examples]

Minimal Working Example

from vegasflow import vegas_wrapper
import tensorflow as tf

def integrand(x, **kwargs):
    """ Function:
       x_{1} * x_{2} ... * x_{n}
       x: array of dimension (events, n)
    """
    return tf.reduce_prod(x, axis=1)

dimensions = 8
iterations = 5
events_per_iteration = int(1e5)
vegas_wrapper(integrand, dimensions, iterations, events_per_iteration, compilable=True)

Please feel free to open an issue if you would like some specific example or find any problems at all with the code or the documentation.

Citation policy

If you use the package please cite the following paper and zenodo references:

• https://doi.org/10.5281/zenodo.3691926
• https://arxiv.org/abs/2002.12921

    @article{Carrazza:2020rdn,
        author = "Carrazza, Stefano and Cruz-Martinez, Juan M.",
        title = "{VegasFlow: accelerating Monte Carlo simulation across multiple hardware platforms}",
        eprint = "2002.12921",
        archivePrefix = "arXiv",
        primaryClass = "physics.comp-ph",
        reportNumber = "TIF-UNIMI-2020-8",
        doi = "10.1016/j.cpc.2020.107376",
        journal = "Comput. Phys. Commun.",
        volume = "254",
        pages = "107376",
        year = "2020"
    }


    @software{vegasflow_package,
        author       = {Juan Cruz-Martinez and
                        Stefano Carrazza},
        title        = {N3PDF/vegasflow: vegasflow v1.0},
        month        = feb,
        year         = 2020,
        publisher    = {Zenodo},
        version      = {v1.0},
        doi          = {10.5281/zenodo.3691926},
        url          = {https://doi.org/10.5281/zenodo.3691926}
    }