Skip to content

GSoC 2020 Kris Akira Stern

Jump to bottom Edit New page
Nabil Freij edited this page Feb 22, 2024 · 4 revisions

glue-solar

Organization: OpenAstronomy (Sunpy)

Requirements: Python, Qt or other GUI programming experience

Section 1: Applicant as a Person

Applicant (Full) Name: Kris Akira STERN

Current University Enrolments:

  • University of London (London, U.K.), B.Sc. in Computing and Information Systems, Year 2
  • University of Hong Kong (Hong Kong SAR, China), Ph.D. in Physics (Observational Astrophysics), Year 4

Points of Contact:

Short Bio:

I am currently a university student pursuing both a graduate Ph.D. degree in Observational Astrophysics in Hong Kong at the University of Hong Kong, and also a B.Sc. degree in Computing and Information Systems by distance at the University of London (London, U.K.) concurrently. My academic interests lie in the stellar evolution of evolved stars of low-to-intermediate-mass stars during the planetary nebula phase, applying nebular diagnostics as a means to study the processes through which the planetary nebula recycles back to the intermediate stellar medium (ISM), serving as a breeding ground for new stars as the planetary nebula expands and cools. I am also an open-source software enthusiast, and am particularly drawn to both Python and C++ programming, though I am slightly stronger in the former. Previously I was chosen as a GSoC 2019 Student Developer, and was fortunate to be working with the OpenAstronomy sub-organization Sunpy to develop the formerly IRISpy project (now known as the sunraster project) with mentors Dr. Dan Ryan and Dr. Laura Hayes. The project has been completed successfully. More recently, I have some working experience in the industry as a web developer for a startup, so am proficient in HTML/CSS and Javascript as applied in the React Javascript framework. My plans for the future is to either become a software engineer working in academia or industry, or to work as some type of researcher in academia engaging in research in some physics/astrophysics discipline. Previously, I have also been receiving training in Deep Learning and Computer Vision through Udacity's Nanodegree programs offered in 2019/20. Also, I was recently an Intel® Edge AI Scholarship recipient, and have received some very basic training in using Intel's OpenVINO toolkit for Internet of Things (IoT) development. In my spare time, I am also an occasional Kaggle dabbler, and am interested in some science-related applications of artificial intelligence (AI).

Section 2: Applicant as a Programmer

  • What is your experience programming? Tell us about something you have created.

I have over one year of open-source programming experience, mostly with Astropy (including Learn Astropy). Computing-language wise, I have roughly five years of programming experience in Python, about one year in C/C++, four years in HTML/CSS, Javascript, PHP for web development, and one year of solid experience with GitHub. Something that I have helped created so far are some source code to the IRISpy project, now “sunraster”, can be accessed at https://summerofcode.withgoogle.com/archive/2019/projects/6094580905148416/.

  • What is your experience with Python? What are your favorite features of Python that are lacking in most other common programming languages? What, in your opinion, is the most advanced Python language feature or standard library functionality that you have used?

I have been a Python programmer contributing to open-source projects for a little over a year since about January 2019. During this time, I have tried my hands on the Sunpy, Astropy, and PySyft projects. I have also worked on data science projects using Python, participating in several projects on Kaggle in order for me to gain some baseline level intuition in the discipline. Moreover, I have successfully completed the Deep Learning and Computer Vision Nanodegrees offered by Udacity in late 2019/early 2020. Python is a high-level language, so its syntax is very effective yet language-like. However, due to its being high-level, it tends to be very slow compared to some low-level languages such as C/C++ that is preferred for low-latency projects. The best feature about Python is the multitude of packages and libraries available. I particularly like Numpy and Pandas, which offers us to have numpy arrays and pandas dataframes which are very useful in storing structured data via their data structures, that may seem unnatural in some other languages such as Java or Javascript. In my opinion, the most advanced Python language feature or standard library functionality that I have used are the lambda function as well as the “map”, “reduce”, and “filter” methods. When used properly, these advanced features can simplify the code a lot.

  • Have you ever used git or another version control system?

Yes, I have used git for version control and am quite proficient in it, both for my open-source contributions and for work. I am still used to using the command line interface via the terminal to make commits to git. But I also know how to use the shortcuts provided by tools such as VS Code and SourceTree (a free git GUI or Mac and Windows) for version control on git as well. I am also acquainted with Apache Subversion for software versioning.

Section 3: Applicant and the Project Idea

  • What do you want to achieve?

Through the proposed GSoC journey, I would like to hone my skills as a software engineer, and perhaps even using it to pave a way for my entrance into the solar physics research field. I have worked as a GSoC Student Developer once before, and really enjoyed my experience with Sunpy in the summer of 2019. I would like to continue on the strength of my previous programming work in related solar physics work on IRISpy using Python to contribute further to the solar physics community. Hopefully my improvements throughout the past year leading up to the project will show, and I will be able to manage to produce some software of great UX/usability to aid multidimensional solar physics data visualization. If that goal can be accomplished it would be like a dream come true for me.

  • What excites you about this project? Why did you choose it?

What excites me most about the project is the prospect of working with Python for programming tasks, with Python being one of my most favorite programming languages. Another aspect of the project idea that is particularly enticing to me is the nature of the glue-solar plugin to be produced, since I am really passionate about data visualization in general, and especially its potential use in solar physics. However, the most important factor I am considering when choosing the project idea are the proposed mentors for the project. I have been fortunate enough to have at least met all three mentors on previous occasions, through my previous contributions to both Sunpy and Astropy. It is my firm belief that the quality of mentorship will be crucial as the make-or-break factor of the project outcome.

  • What have other people done on this idea? Has it been implemented before?

Technically speaking the idea has been explored and applied to other disciplines for data visualization. The project uses Glue, which is a “python project to link visualizations of scientific datasets across many files.” What this project is concerned about is a specific plugin of Glue for solar physics that uses Sunpy. Similar plugins have been developed for use in other disciplines, such as glue-medical for medicine, glue-geospatial for the geosciences, and glue-wwt, glue-aladin, glue-samp, specviz, as well as STScI's cubeviz and mosviz for astronomy (see: http://docs.glueviz.org/en/stable/customizing_guide/available_plugins.html), but none for solar physics specifically exists thus far. Therefore, there will be ample examples we will be able to use as references if the glue-solar project will get selected this year.

  • What qualifications do you have to implement your idea? Why are you suited to work on this project?

I have engaged in some Sunpy programming work on NDCube and the formerly IRISpy last year in 2019 as a GSoC Student Developer which are Python-based, and such experiences will definitely help with the proposed GSoC 2020 work for the months ahead which can serve as a continuation of my journey with Sunpy. Currently IRISpy has been renamed to “sunraster” and is undergoing great transformations, such that much has changed since I last worked on it last summer. Despite that, I am suited to work on this project because it seems like a natural follow-up for me as the glue-solar project idea is more challenging than the IRISpy project idea I worked on last year, and much of what I have learned, including ND cubes and IRIS data, can be applied in the new project. Moreover, for my present academic studies in observational astronomy, I deal with data cubes in the FITS format of planetary nebula a lot of the time with Astropy and other cube-based software such as glue's cubeviz plugin, and am familiar with basic knowledge of astrophysics required to be successful to bring the proposed GSoC project to fruition. I have not done much GUI programming in Python with Qt, but am keen to learn, so that is another motivation for me to be engaged in the project.

  • Some Use Cases of glue-solar

  1. Some testing/improvements are needed for Glue-Solar's beta feature for creating derived datasets from pixel selections, which is more efficient than the normal subset extraction, as follows:

    • To ensure that the subsets extracted with the tool have the correct WCS information and are linked to the other datasets. For example, when highlighting subsets, these should show up in the extracted dataset.
    • To find a good icon for the pixel extraction tool
    • To implement other UX improvements

  2. To enable extraction of a 1D profile from a pixel selection of a dataset with greater than 3 in dimensions (or ND > 3D), and be able to perform animation that would profile along any extra dimensions. For example, taking a (space, space, wavelength, time) cube, select a pixel in (space, space) leading to a 2D derived dataset with the dimensions of (wavelength, time). Currently, Glue will collapse the time dimension, when it would be good for it to be able to provide a slider functionality such that one can see the wavelength profile with time.

  3. Some testing of dataset auto-linking for subsets of ND cubes is warranted. Currently, if one loads in two ND cubes, like (space, space, time) and (space, space, wavelength) there is special handling in the auto-linking code to link the two spatial axes. However, if one loads in a (space, space, time) and a (time, wavelength) dataset the time axes will not be autolinked. This behavior might need to be modified.

  4. To enable extraction of derived datasets from a path, this would allow drawing a line on an image and extracting a derived dataset for the pixels under that line. (Mentor Tom R. has a prototype of this which can be used as a reference.)

  5. To enable Dask support. Especially, to facilitate the not loading of all the array into memory for calculating pixel limits.

  6. To enable image/movie export, both with axes and without axes, via matplotlib.

  7. To develop Solar physics loaders, specifically for NDCube/NDData, SST, IRIS, EIS, DKIST -type data. (With the addition of AIA, HMI data if possible.)

  8. To enable auto selection of the "correct" solar physics colormap(s).

  • Goals to be Reached for the GSoC Project

The following list is arranged in descending order of priority/importance:

  1. Glue-solar as an installable package, with a first official release.
  2. Glue-solar is well documented, both for users and future developers, with screenshots and screencasts as appropriate.
  3. Glue-solar can load many different complex datasets such as NDCube/NDData, SST cubes, IRIS data, EIS data, DKIST data as well as AIA data, HMI data, among many others.
  4. Glue-solar demonstrates how to implement custom data loaders (i.e. for IRIS data).
  5. Glue-solar demonstrates how to implement custom linked layouts (i.e. SJI plus Raster IRIS data).
  6. Glue-solar includes a custom splash screen with the Sunpy logo.
  • Proposed Time Commitment

I am planning to invest between 30 and 40 hours per week in carrying out the project during the entire duration of the coding period (with on average 5 to 6 hours per day, 7 days a week). And, I will take no summer vacation for the entire duration of GSoC 2020. Also, I plan on committing to the project long-term if possible after the completion of GSoC this year, so am prepared to continue work even after the conclusion of the program to ensure the goal of delivering the glue-solar as an installable package, with a first official release can be accomplished in the near future.

  • Proposed Student Timeline

Glue already supports many types of solar data out-of-the-box. Hence the overarching guiding principle in driving this project in particular is to develop a feature-complete version of the software and to focus on edge features on top of the existing functionalities offered by glue. As a reference, the sort of features that exist in other interactive data exploration tools include the following list:

  1. Able to open the files and understand the metadata. (Existing feature)
  2. Creating movies or PNGs of different frames and wavelengths.
  3. Basic slit analysis of images through several dimensions.
  4. Basic Doppler and magnetic field support. (Nice-to-have feature)
  5. Re-scaling images. (Existing feature)
  6. Line profile fitting. (Existing feature)

Phase I: Community Bonding Period (May 5, 2020 - June 2, 2020 HKT)

  • I plan to use such a period to bond with the Sunpy community at large as well as with my mentors, so as to foster good working relationships.
  • Also, I will use this time to carry out detailed research regarding how glue works as a codebase, as well as investigate how other similar plugins are written and perform in order to learn from existing work that can be applied in the glue-solar work. I will go over the references given in this proposal thoroughly during this period, including Goodman et al. (2012), Beaumont et al. (2015), Robitaille et al (2017a), and Robitaille et al (2017b).

Phase II: First 4-weeks (from June 3, 2020 up until the 1st Evaluation during June 30, 2020 - July 4, 2020 HKT)

  • Week 1: Document and test the loading of AIA & HMI data and show built-in alignment with glue-solar. (Carry out unit testing at every step.)
  • Week 2: Test and document re-scaling of images with glue-solar as well as the existing line profile fitting feature for solar physics data.
  • Week 3: Enable glue-solar to create movies or PNGs of different frames and wavelengths.
  • Week 4: Enable glue-solar to carry out basic slit analysis of images through several dimensions.

Phase III: Second 4-weeks (from July 5, 2020 up until the 2nd Evaluation during July 28, 2020 - August 1, 2020 HKT)

  • Week 5: Enable glue-solar to demonstrate how to implement custom linked layouts (i.e. SJI plus Raster IRIS data).
  • Week 6: Enable glue-solar to demonstrate how to implement custom data loaders (i.e. for IRIS data).
  • Weeks 7-8: Enable glue-solar to load many different complex datasets such as NDCube/NDData, SST cubes, IRIS data, EIS data, DKIST data as well as AIA data, HMI data, among many others.

Phase IV: Third 4-weeks (from August 2, 2020 up until the final Evaluation for Students during August 25, 2020 - September 1, 2020 HKT)

  • Week 9: Test and integrate existing prototype of path extraction tool into glue-solar.
  • Week 10: Further develop, test, and document the solar datasets extraction tool, and combine this with the path extraction tool developed previously in Week 9. Upstream the new tool into the glue core if it is sufficiently polished.
  • Week 11: Enable glue-solar to include a custom splash screen with the Sunpy logo, and add basic Doppler and magnetic field support to glue-solar if time permits.
  • Week 12: Complete glue-solar documentation so that the package will emerge as well documented, both for users and future developers, with screenshots and screencasts as appropriate. Carry out beta testing.

Phase V: Post-GSoC (September 2, 2020 HKT and beyond)

  • Work further on completing the documentation for the package.
  • Work further on adding more features to the package, such as adding Dask support to glue-solar.
  • Work further towards finalizing the package for first official release.

Section 4: Links to Patches

During the recent months (within the past six months), the following patches have been submitted as my personal open-source contributions, the first of which is for satisfying the basic patch requirement of Open Astronomy.

References

  1. Goodman et al. (2012), Principles of high-dimensional data visualization in astronomy
  2. Beaumont et al. (2015), Hackable User Interfaces In Astronomy with Glue
  3. Robitaille et al (2017a) glueviz v0.10: multidimensional data exploration
  4. Robitaille et al (2017b) glueviz v0.13.1: multidimensional data exploration

Recorded in the Wiki on April 5th, 2020.

Add a custom footer

SunPy

Clone this wiki locally