GSoC 2020 Report: Upgrade and Fix tvb-gdist C++ Library

Thank You Note

It was a pleasure working under my mentors Lia and Paula. Thank you for the constant guidance and valuable feedback whenever I was stuck. I express my heartfelt gratitude to Lia, Paula, the whole The Virual Brain and INCF team and, Google for making the last four months such a rewarding experience!

About

While running simulation on cortical surfaces we need to calculate geodesic distance as opposed to euclidean distance due to the shape of the cortical surface. The virtual brain uses geodesic_library for this calculation. The library implements the original paper in C++. The original source code can be found in Google Code Archive: https://code.google.com/archive/p/geodesic. tvb_geodesic repository implements a cython wrapper on top of the C++ code which then is released to Pypi (tvb-gdist) and conda-forge (Tvb Gdist).

However, the code is now outdated and users have reported various issues. In this project, we aim to update the code and fix those issues.

In this project, we fixed the 5 existing issues. Along with that, we fixed issues reported by users during the GSoC period (excessive memory usage, issue #51), added new API (issue #52), added a documentation (issue #43) so that users do not need to go through the code and made the code more maintainable by adding tests, implementing coverage reporting (issue #57) and following lint conventions (issue #60). A detailed descriptions of work done is mentioned below.

Usage

The package exposes 3 APIs which can be used for various purposes: compute_gdist, local_gdist_matrix and distance_matrix_of_selected_points. Examples and detailed instructions can be found in the project readme page.

Project Status

• Completed Work

  • Fixing an issue where numpy requirement is causing installation problems using pyproject.toml
  • Lint fixes, the codebase is now PEP8 compliant
  • Adding coverge report functionality
  • Updating the travis CI configuration
  • Fixing memory leak issue for compute_gdist
  • Adding a new API for pairwise distance computation
  • Adding documentation in the README file
  • Running different versions of Python on Windows
  • Enabing users to specify if a the faces vertices’ are 1-indexed
  • Adding googletest to test and measure coverage of the c++ library
  • Adding parallelism using OpenMP

• Incomplete Work

  • Using ctypes instead of cython

Links

• Link to the project in GitHub
• Link to the project in PyPI
• Link to the list of commits
• Link to the pull requests
• Link to the weekly reports
• Link to Jira story

List of Commits