Metadata-Version: 2.1
Name: parallel-delaunay
Version: 0.1.0
Summary: Parallel implementation of Guibas & Stolfi's divide-and-conquer algorithm for Delaunay triangulation, using MPI in Python.
License: MIT
Author: v-hill
Author-email: 67915924+v-hill@users.noreply.github.com
Requires-Python: >=3.7,<4.0
Classifier: License :: OSI Approved :: MIT License
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.7
Classifier: Programming Language :: Python :: 3.8
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Requires-Dist: matplotlib (>=3.6.2,<4.0.0)
Requires-Dist: mpi4py (>=3.1.4,<4.0.0)
Requires-Dist: numpy (>=1.23.5,<2.0.0)
Requires-Dist: scipy (>=1.10.0,<2.0.0)
Description-Content-Type: text/markdown

# delaunay-triangulation
This library is used for computing the [Delaunay triangulation](https://en.wikipedia.org/wiki/Delaunay_triangulation) of 2D point sets.

An example Delaunay triangulation output for 32 points is shown below.
<img src="./images/Figure 2021-02-24 110617.png" alt="drawing" width="500"/>

**Why is this library special?**

Speed! This is a pure python implementation of the Guibas &amp; Stolfi's divide and conquer algorithm. The divide and conquer algorithm has been shown to be the [fastest](https://people.eecs.berkeley.edu/~jrs/meshpapers/SuDrysdale.pdf) DT generation technique, with O(*n* log *n*) running time. Furthermore, this code has been parallelised using the MPI for Python ([mpi4py](https://github.com/mpi4py/mpi4py)) library to utilise multiple CPU cores. This allows the algorithm to be efficiently scaled for distributed computing across supercomputer nodes.

# Setup
1. Install anaconda or miniconda
2. Install git, then clone repository: `git clone https://github.com/v-hill/delaunay-triangulation`
3. Create environment: `conda create -n deltri python=3.9`
4. Activate environment: `conda activate deltri`
5. Install dependencies: `conda env update -f conda_env.yml --prune`
6. Run test: `python src/triangulation_test.py`

# Benchmarking

### Single core results

As seen in the graph below, this Delaunay triangulation implementation scales just as efficiently as the SciPy implementation. The SciPy library is written in C and wrapped in Python for ease of use. As a result, the SciPy DT algorithm is consistently around 40x faster for a given number of points. However, the benefit of my library is that it can take advantage of multiple CPU cores, offering a performance advantage unavailable to SciPy users. By utilising the multiple cores available in modern computers, the run time can be reduced linearly by a factor approximately equal to the number of available threads.

<img src="./images/Figure 2021-07-02 210145.png" alt="drawing" width="700"/>

This graph can be reproduced by running `python src/triangulation_benchmarks.py`.

## Structure
This repository is currently structured as follows.

    ├── src
        ├── triangulation_core
            ├── points_tools
                ├── generate_values.py
                └── split_list.py
            ├── edge_topology.py
            ├── linear_algebra.py
            ├── triangulation.py
            └── triangulation_primitives.py
        ├── utilities
            └── settings.py
        ├── triangulation_benchmarks.py
        ├── triangulation_cli.py
        ├── triangulation_mpi_test.py
        └── triangulation_test.py

## References
<a id="1">[1]</a>
Guibas, Leonidas and Stolfi, Jorge
'Primitives for the Manipulation of General Subdivisions and the Computation of Voronoi'
In: ACM Trans. Graph.4.2  (Apr.1985),  pp.  74–123.
issn:  0730-0301
doi:10.1145/282918.282923

