{"author":"Ivan Karpukhin","author_email":"karpuhini@yandex.ru","bugtrack_url":null,"classifiers":[],"description":"# Batch linear assignment for PyTorch\n[![PyPI version](https://badge.fury.io/py/torch-linear-assignment.svg)](https://badge.fury.io/py/torch-linear-assignment)\n[![Build Status](https://github.com/ivan-chai/torch-linear-assignment/actions/workflows/ci-tests.yml/badge.svg)](https://github.com/ivan-chai/torch-linear-assignment/actions)\n[![Downloads](https://static.pepy.tech/badge/torch-linear-assignment)](https://pepy.tech/project/torch-linear-assignment)\n[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)\n

\n
\n Installation |\n Usage |\n Citation\n
\n

\nBatch computation of the linear assignment problem on GPU.\n\n## Install\nBuild and install via PyPI (source distribution):\n```bash\npip install torch-linear-assignment\n```\n\nBuild and install from Git repository:\n```bash\npip install .\n```\n\nWhen building with CUDA, make sure NVCC has the same CUDA version as PyTorch.\nYou can choose CUDA version by\n```bash\nexport PATH=/usr/local/cuda-/bin:\"$PATH\"\n```\n\nIf you need custom C++ compiler, use the following command:\n```bash\nCXX= CC= pip install .\n```\n\nIf you get a torch-not-found error, try the following command:\n```bash\npip install --upgrade pip wheel setuptools\npython -m pip install .\n```\n\n## Example\n```python\nimport torch\nfrom torch_linear_assignment import batch_linear_assignment\n\ncost = torch.tensor([\n 8, 4, 7,\n 5, 2, 3,\n 9, 6, 7,\n 9, 4, 8\n]).reshape(1, 4, 3).cuda()\n\nassignment = batch_linear_assignment(cost)\nprint(assignment)\n```\n\nThe output is:\n```py\ntensor([[ 0, 2, -1, 1]], device='cuda:0')\n```\n\nTo get indices in the SciPy's format:\n```py\nfrom torch_linear_assignment import assignment_to_indices\n\nrow_ind, col_ind = assignment_to_indices(assignment)\nprint(row_ind)\nprint(col_ind)\n```\n\nThe output is:\n```py\ntensor([[0, 1, 3]], device='cuda:0')\ntensor([[0, 2, 1]], device='cuda:0')\n```\n\n# Citation\nThe code was originally developed for the [HoTPP Benchmark](https://github.com/ivan-chai/hotpp-benchmark). If you use this code in your research project, please cite one of the following papers:\n```\n@article{karpukhin2024hotppbenchmark,\n title={HoTPP Benchmark: Are We Good at the Long Horizon Events Forecasting?},\n author={Karpukhin, Ivan and Shipilov, Foma and Savchenko, Andrey},\n journal={arXiv preprint arXiv:2406.14341},\n year={2024},\n url ={https://arxiv.org/abs/2406.14341}\n}\n\n@article{karpukhin2024detpp,\n title={DeTPP: Leveraging Object Detection for Robust Long-Horizon Event Prediction},\n author={Karpukhin, Ivan and Savchenko, Andrey},\n journal={arXiv preprint arXiv:2408.13131},\n year={2024},\n url ={https://arxiv.org/abs/2408.13131}\n}\n```\n","description_content_type":"text/markdown","docs_url":null,"download_url":null,"downloads":{"last_day":-1,"last_month":-1,"last_week":-1},"dynamic":["Author","Author-Email","Description","Description-Content-Type","Requires-Dist","Summary"],"home_page":null,"keywords":null,"license":null,"license_expression":null,"license_files":null,"maintainer":null,"maintainer_email":null,"name":"torch-linear-assignment","package_url":"https://pypi.org/project/torch-linear-assignment/","platform":null,"project_url":"https://pypi.org/project/torch-linear-assignment/","project_urls":null,"provides_extra":null,"release_url":"https://pypi.org/project/torch-linear-assignment/0.0.3/","requires_dist":["torch>=1.12.0","scipy>=1.6.0"],"requires_python":null,"summary":"Batched linear assignment with PyTorch and CUDA.","version":"0.0.3","yanked":false,"yanked_reason":null}

\n \n Installation |\n Usage |\n Citation\n \n

\n
\n Installation |\n Usage |\n Citation\n
\n

Dependencies of torch-linear-assignment (2).
Dependency	Optional
scipy	false
torch	false

Package search

torch-linear-assignment

project_urls

Dependencies

All package metadata

Releases