Skip to content

GSoC 2016 Application Anchit Jain:Implement Public API for ERFA

Jump to bottom Edit New page
Anchit Jain edited this page Apr 8, 2016 · 4 revisions

Student Info

Name Anchit Jain
Email anchitjain1234@gmail.com
Github anchitjain1234
IRC Nick anchitjain@freenode
Time Zone IST (UTC + 05:30)
GSOC Blog URL http://www.anchitja.in/blog/
Blog RSS feed http://www.anchitja.in/atom.xml/
University information

Code Sample

I have submitted following patches to astropy

Merged :
  1. #4606 - Support for Path objects in io packages.
  2. #4561 - Remove new line characters after last row of data in ascii.latex.AASTex.
  3. #4532 - Fix to generate zero-length copy of current table.
  4. #4482 - Displaying Compound model expressions when printing compound model instances.
  5. #4474 - Fix for CDS reader requiring at least two description characters.
Under review :
  1. #4585 - Error handling in io.fits.compression module.
  2. #4574 - Allow passing of tuples for setting constraints on model parameters.
  3. #4551 - Explicitly set BeautifulSoup parser to html.parser to avoid warning.
  4. #4505 - Fix for allowing WCS to take Image HDU.

Project Info

  • Project Title

    Implement Public API for ERFA

  • Project Abstract

    Astropy currently uses wrapped version of ERFA library which covers some functions from original C library but not all.
    If this wrapped library is extended to wrap all functions under ERFA and exposed as a public API, this would help astropy to leverage a validated and trusted library to do computational "heavy-lifting" which might also lead to performance improvements.
    The two major aims of this project are

    1. Create a separate package under liberfa which would be having 100% coverage for ERFA functions, good test coverage and documentation such that it follows all the guidelines for astropy affiliated package.
    2. Use the new package in astropy replacing _erfa and discuss for replacing overlapping code between astropy python code and ERFA.

  • Deliverables

    1. Working python based ERFA package wrapping original C based ERFA package, strictly adhering to astropy affiliated package guidelines.
    2. Removal of astropy._erfa and use of this new python based ERFA package as external (but bundled) dependency.

  • Proposal Detailed Description

    ERFA is Essential Routines for Fundamental Astronomy which implements high-quality astronomical routines in C.
    Currently astropy uses ERFA in coordinates and time subpackages for some angle utilities, conversions and time conversion, formatting by using _erfa subpackage.
    But it has been under discussion in community for a long time (#3123, #4571) to use ERFA as an external package in astropy by exposing ERFA as public API. Recently (#4664) there has been also some discussion to whether replace python code in coordinates subpackage with ERFA code.
    Therefore exposing ERFA as public API would be beneficial for Astropy and the larger scientific Python community.
    I propose to do the following changes in two phases:

    1. Phase 1(7 weeks) - ERFA as separate package

      Right now for ERFA functionality, astropy uses _erfa subpackage which wraps ERFA functions only from Astronomy section, AngleOps and SphericalCartesian subsections under VectorMatrix section.
      In this phase I propose to increase wrapping and test coverage, improve function names and documentation.
      I intend to do this in 3 subphases:

      1. Extension of wrapping
        • First I would start with Astropy affiliated package template and using the existing _erfa code in it, to create a new package possibly named pyerfa. Then I intend to extend it(wrapping) to all the remaining sections/subsections such that package could be imported as independent package, and all the ERFA functions would work without worrying about correctness of the results, just get every ERFA function working.
          • Currently functions under CopyExtendExtract under VectorMatrix do give some weird error on wrapping. So my first task would be to parse functions under CopyExtendExtract correctly.
          • Functions under other subsections were getting compiled correctly so next task would be to check that they are parsed correctly i.e. check if they do work and if not correct them.
          • Existing docstring parser would need to be modified such that new parts to the docstrings could be added which are not present currently in original C library.
      2. Testing of wrappings
        • In this subphase I intend to test rigorously all the new wrappings extended to ensure result correctness following Astropy guidelines for testing.
        • This would involve testing the wrappings under different values of arguments and input types etc. using py.test module and following Astropy standard for testing.
        • Existing C test suite for ERFA (t_erfa_c.c) would be ported to Python by automatic conversion in the similar manner, by creating parsing templates for test suite and test suite generator. Parts which couldn’t be ported, would be hand-written. Through this approach we would be having exactly same test suite as of original SOFA/ERFA tests.
        • Overall objective would be to create a test suite having coverage >= 90%.
      3. Create name aliases and improve documentation:
        • In this subphase I would work on creating name aliases for the functions from ERFA as current names are very cryptic which don't suit well for code readability and understanding. For example. eraAb is a function in ERFA that performs aberration to transform natural direction into proper direction but this isn't very much clear from this name. Documentation improvement would go hand in hand with name aliasing
        • Name aliasing would involve manual work of checking code, work to be done by function, docs, arguments for function etc. and most importantly discussion with mentors to be done for every function.
        • Documentation improvement would include updating of current docstrings generated by C comments to include examples following doctest, short summary and other notes to follow numpy style docstrings.
    2. Phase 2(5 weeks) - ERFA as separate package

      Under this phase my focus would be on to use this newly generated package within Astropy.
      I intend to do this in 2 subphases:

      1. Update _erfa usage
        • As _erfa would no longer be needed therefore update usages of _erfa by the new package and get everything working
        • This would be done by creating new ERFA package as external dependency by placing it in astropy/extern and update its usages. Then _erfa would be removed.
      2. Overlapping code replacement
        • This subphase would involve discussion to whether replace python code with ERFA based on the results of performance analysis.
        • Here first I would try to find parts of code where ERFA could be used, mainly in coordinates, time etc. This would be done by carefully looking at the existing codebase to check whether ERFA function would suit here.
        • Then using performance analysis and discussion with mentors replace the code for the parts where agreement comes.
        • Performance analysis would be done using asv by writing benchmarks first for existing code and then for code using ERFA.

Timeline

Week Work
Community bonding period (22 Apr - 22 May) Check astropy._erfa and liberfa/erfa code closely, discuss concrete approach with mentors, read documentation for asv, cython .
Week 1 (23 May - 29 May)
  • Setup new repository using astropy affiliated package guidelines and _erfa existing code.
  • Make this repository work independently without Astropy.
Week 2 (30 May - 5 June)
  • Extend to CopyExtendExtract subsection.
  • Modify parsing templates.
Week 3 (6 June - 12 June)
  • Extend to new sections of ERFA.
  • Make whole package working.
Week 4 (13 June - 19 June) Port original test suite(t_erfa_c.c) to Python is similar manner (automated like wrapping).
Week 5 (20 June - 26 June)
  • Convert manually tests which couldn't be converted by automation.
  • Ensure good coverage.
Weeks 6-7 (27 June - 10 July)
  • Create name aliases after discussion with mentors.
  • Improve documentation to Astropy standard.
Week 8 (11 July - 17 July)
  • Use new ERFA python wrappings package.
  • Remove existing _erfa subpackage from astropy.
Week 9 (18 July - 24 July ) Check carefully astropy code, mainly in sections coordinates and time where existing python code could be replaced by ERFA function.
Weeks 10-11 (25 July - 7 August) Write benchmarks to evaluate performance of existing code and code after replacing with ERFA.
Week 12-13 (8 August - 23 August)
  • Discuss with mentors after performance analysis which parts to replace and which not.
  • Complete all the remaining spilled work from previous phases.

Other Commitments

  1. Do you have any other commitments during the main GSoC time period, May 23rd to August 23rd?

    • My end semester examinations would be from 4 May - 15 May, so I would not be able to do much research then.
    • My new semester would be starting from 1 August, this might decrease my productivity, but I would try my best to not let it affect me.

  2. Have you applied with any other organizations?
    No. I have just applied to Astropy.

  3. Have you participated previously in GSoC? when? with which project?
    No. This is first time I am applying for GSoC.

Extra information

Link to Resume http://www.anchitja.in/assets/resume.pdf/

Alternate contacts

Add a custom footer
Add a custom sidebar
Clone this wiki locally