Skip to main content

Utilities for the CG:SHOP 2020 Optimization Competition on the Minimum Convex Partition Problem.

Project description

Official Python Utilities for the CG:SHOP 2020 Optimization Competition.

We provide basic code to ease your participation in this year's challenge. Due to it popularity and simplicity, the choice has fallen on Python. However, the real code is written in C++ with CGAL to provide the needed accuracy and efficiency.

This python module allows you to easily read the instance and potentially convert them into an easier format. The JSON format is not as simple as last year's raw format but it allows to add metadata. Further, it allows you to verify your instances. It uses the same core as the server so if this code accepts your solution, so will our server.

The code is not perfect but we will work to improve this code during the competition. Feedback is welcome.

The C++-core unfortunately messes with the platform independence of Python. At this point, we already spent much more time into making the module as compatible as possible possible than into the actual code. Unfortunately, there are still a lot of ugly workarounds, which we try to remove soon. The module has been tested on OS X and Linux. Windows is still experimentally.

We implemented a fallback mode that in case the native core could not be loaded at least the basic functionality for reading and writing solutions and instances is available.

Please report any bugs you may find. You can simply write a mail to cgshop-admin@ibr.cs.tu-bs.de and one of our developers will take care of it (also if you have questions or suggestions).

https://gitlab.ibr.cs.tu-bs.de/alg/cgshop2020_pyutils

Features

  • Reading and writing instance files.
  • Reading and writing solution files.
  • Verifying solutions for feasibility.
  • Simple triangulation solver to get a trivial feasible solution.

The exact same implementation will be used on the submission server. Thus, if this library will accept your solution, so will the submission server.

Installing

You need to have a 64bit operating system (Linux, OS X, Windows) and a 64bit Python interpreter for full functionality. Unfortunately, it is very easy on Windows to accidentally install outdated 32bit versions. 32bit users will only be able to use a limited functionality (or have to compile the core themselves).

The module is available through pypi, e.g.,

pip install cgshop2020-pyutils

or when using the generally recommended pipenv.

pipenv install cgshop2020-pyutils

Note that you have to use _ instead of - when importing the module.

You can also simply copy the folder cgshop2020_pyutils into your code folder and use it like a local module.

If you want to work in C++ directly, you can also simply copy our cpp_core. If you have some experience with CMake, this should be straightforward. However, you will be missing the JSON functionality as this is written in pure Python (which is much simpler than doing this in C++).

To update to the newest version, you have to use

pip install cgshop2020-pyutils --upgrade

or

pipenv update cgshop2020-pyutils

Getting started

With the module installed, you can now try to create your first solution set (simple triangulations) as follows:

# Load an instance
from cgshop2020_pyutils import InstanceDatabase, BestSolutionSet, SolutionZipWriter, TrivialTriangulationSolver

# load challenge instances
idb = InstanceDatabase("/path/to/extracted/zip/")

# compute the triangulation for all instances
triangulation_solver = TrivialTriangulationSolver()
solutions = BestSolutionSet()
for instance in idb:
    solutions.add(triangulation_solver(instance))
    print(f"Computed triangulation for {instance.name}")

# write solutions into zip
print("Creating zip. This can take some time...")
with SolutionZipWriter("my_first_upload.zip") as zipper:
    zipper.add_solutions(solutions)

print("You can now upload 'my_first_upload.zip' on",
      "https://cgshop.ibr.cs.tu-bs.de/competition/cg-shop-2020/")

You can find further examples in ./examples

Instances

You can easily access the competitions instances by downloading the zip-archive of the first instances batch, extracting it, and using the InstanceDatabase.

The InstanceDatabase does only need the directory of the extracted zip and will recursively search it for instances of a specific name, e.g.

from cgshop2020_pyutils import InstanceDatabase
instance_database = InstanceDatabase("/path/to/extracted/zip")
mona_lisa_instance = instance_database["mona_lisa_1000000"]

alternatively, you could also simply iterate over all instances:

from cgshop2020_pyutils import InstanceDatabase
instance_database = InstanceDatabase("/path/to/extracted/zip")
for instance in instance_database:
    print(f"Found instance {instance.name}")

The instance class is rather simple:

from cgshop2020_pyutils import InstanceDatabase
instance_database = InstanceDatabase("/path/to/extracted/zip")
instance = instance_database["uniform-0000100-2"]
print(f"The instance is named {instance.name}.")
print(f"It has {len(instance.name)} points.")
print(f"The first point is f{instance[0]}.")
print("I now list the other points:")
for point in instance:
    print(f"({point.get_x()}, {point.get_y()}")

You could also create your own instance by

from cgshop2020_pyutils import Instance, Point
instance = Instance(name="my_instance", points=[Point(1.0, 1.0), Point(0.0, 3.0), Point(2.0, 0.0)])
instance.add_point(Point(1.0, 5.0)) # add an additional point

Instance Format

An instance in json can look likes this:

{"points": [
            {"i": 0, "x": 2396.0, "y": 5284.0}, 
            {"i": 1, "x": 2656.0, "y": 2938.0}, 
            {"i": 2, "x": 4120.0, "y": 2278.0}, 
            {"i": 3, "x": 4342.0, "y": 102.0}, 
            {"i": 4, "x": 4384.0, "y": 2988.0}, 
            {"i": 5, "x": 5136.0, "y": 2280.0}, 
            {"i": 6, "x": 6634.0, "y": 5416.0}, 
            {"i": 7, "x": 8598.0, "y": 2632.0}, 
            {"i": 8, "x": 8898.0, "y": 4170.0}, 
            {"i": 9, "x": 11738.0, "y": 1550.0}
           ], 
  "type": "Instance", 
  "name": "euro-night-0000010", 
  "meta": {
            "comment": "HIP even point set instance (10 points) sampled from image ", 
            "faces_in_triangulation": 12
           }
}
  • points: a list of points with (i=index, x=x coordinate, y=y coordinate)
  • type: "Instance", just tells you that this json describes an instance
  • meta some optional metadata
    • comment: Some comment on the instance
    • faces_in_triangulation: Number of faces in the delaunay triangulation as trivially achievable objective value.
  • name: name of instance.

Last time we used some raw text format which is easier to parse but less powerful. If you don't want to read json in your software, we recommend to convert the instances via our python library to a simpler format.

from cgshop2020_pyutils import InstanceDatabase
instance_database = InstanceDatabase("/path/to/extracted/zip")
for instance in instance_database:
    with open(f"{instance.name}.txt", "w") as simple_file:
        for index in range(len(instance)):
            point = instance[index]
            simple_file.write(f"{index} {point.get_x()} {point.get_y()}\n")

For loading an instance from a specific file, you can use the InstanceReader.

Solution

from cgshop2020_pyutils import Solution, Edge
solution = Solution(instance="my_instance")
solution.add_edge(Edge(0, 1)) # edge from point with index 0 to point with index 1
solution.add_edge(Edge(0, 2)) # edge from point with index 0 to point with index 2

solution.delete_double_edges() # remove redundant edges, as double edges are illegal.
for edge in solution:
    print(edge)
for i in range(len(solution)):
    edge = solution[i]
    print(edge)

you can easily create a valid json file:

from cgshop2020_pyutils import SolutionWriter
writer = SolutionWriter()
writer.to_json(solution=solution, path="my_solution.json")

to create a zip for upload, use

from cgshop2020_pyutils import BestSolutionSet, SolutionZipWriter

solutions = BestSolutionSet()
solutions.add(solution)
with SolutionZipWriter("my_upload.zip") as zipper:
    zipper.add_solutions(solutions)

We will probably add a functionality for automatic upload from code in the near future.

Meta data

Solution and Instance both have a meta_data attribute which is a dictionary. For example you can add a comment by

my_solution.meta_data["comment"] = "This is a comment"

which will also be saved when converting to JSON. Everything that can be converted to string will be saved to json and also reloaded when reading the json.

Checker

We provide an efficient checker for verifying solutions. In case of infeasibility, the checker also provides a message with an error description. If the solution is feasible, the objective value is computed.

from cgshop2020_pyutils import SolutionChecker

checker = SolutionChecker()
status = checker(instance=instance, solution=solution)
print(status.is_feasible())
print(status.get_message())
print(status.get_objective_value())

Visualization

We provide a simple visualization based on matplotlib to quickly plot instances and solutions.

from cgshop2020_pyutils import Visualizer
vis = Visualizer()
vis.visualize_solution(solution=solution, instance=instance) # opens plot if possible
vis.visualize_solution(solution=solution, instance=instance, path="my_fig.pdf") # writes plot to file

Trivial Triangulation Solver

We provide a simple solver that just computes the Delaunay triangulation.

from cgshop2020_pyutils import TrivialTriangulationSolver
solver = TrivialTriangulationSolver()
solution = solver(instance)

Check a zip

from cgshop2020_pyutils import ZipSolutionIterator, SolutionChecker, InstanceDatabase
checker = SolutionChecker()
idb = InstanceDatabase("./my_folder_with_the_instances")
zsi = ZipSolutionIterator() # Also allows to set some constraints on file size, etc.
for solution in zsi("myzip.zip"):
    instance = idb[solution.instance_name]
    assert checker(instance=instance, solution=solution).is_feasible()

Compiling yourself

You should not need to do this as the module comes with precompiled binaries. However, it is possible to recompile the C++-core on your machine.

You need to have the following libraries installed:

  • A C++ compiler e.g. Debian/Ubuntu apt-get install g++
  • CMake Debian/Ubuntu: apt-get install cmake, OS X:brew install cmake
  • CGAL Debian/Ubuntu: apt-get install cgal, OS X:brew install cgal
  • Boost Debian/Ubuntu: apt-get install boost, OS X:brew install boost
  • Python3 (of course)
  • matplotlib in Python3 (e.g., pip3 install matplotlib)

Otherwise, you likely get an error.

Do a manual recompile with:

import cgshop2020_pyutils
cgshop2020_pyutils.compile_cpp_core()

This will only work if you installed into user space (without sudo) as otherwise the compiler cannot write the files into the module folder.

This command actually just executes CMake. You could also do this by hand.

For Windows, CMake might use slightly different paths. In that case, you could simply copy the created .dll into the binaries folder.

Changelog

  • 0.1.0 First published version with native support for OS X and Linux.
  • 0.1.1 Experimental Windows support
  • 0.1.2 32bit detection and warning.
  • 0.1.3 fixed a problem with some Windows versions (could not find the supplied dependencies in the same folder)
  • 0.1.4 Implemented InstanceDatabase, fixed bug with reading solution (thanks to Johannes Obenaus)
  • 0.1.5 Bugfix in InstanceDatabase
  • 0.2.0 Zip reading utilities.
  • 0.2.1 Missing dependency declaration in pip package (chardet).
  • 0.3.0 Added zip writing utilities and examples.
  • 0.3.1 Removed potential problem in InstanceDatabase, improved documentation. Fixed a bug in zip reader (which makes me wonder why it worked on the test server?)

Roadmap/Planned features

  • Proper testing with 100% coverage (currently don't have the manpower for that)
  • Zipping the solution sets
  • Upload of zips
  • Fetching the objective value for your currently best solutions from the server (and skipping upload of worse solutions)

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

cgshop2020_pyutils-0.3.2-py3-none-any.whl (8.8 MB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page