Skip to content

GSoC 2021 Adwait Bhope

Jump to bottom Edit New page
Nabil Freij edited this page Feb 22, 2024 · 6 revisions

Adding Support for Resampling Data with NDCube

Organization: OpenAstronomy Sub-organization: sunpy

Mentors:

  1. Dan Ryan (@DanRyanIrish)
  2. Stuart Mumford (@Cadair)

Personal Details

Adwait Bhope adwaitbhope@gmail.com

GitHub: @adwaitbhope Matrix: @adwaitbhope

Final year CS undergrad at University of Pune(SPPU), admitted to Pune Vidyarthi Griha's College of Engineering and Technology, India.

Google Summer of Code

  • Have you participated in GSoC previously? No, this is the first time I'm applying to an organization.

  • Are you also applying to other projects? Yes, I plan to apply to another one of sunpy's projects: Update sunraster to ndcube 2.0 under OpenAstronomy.

  • Are you eligible to receive payments from Google? Yes, I'm eligible.

  • How much time do you plan to invest in the project before, during, and after the Summer of Code? I will be able to satisfy the official GSoC guideline of 18 hours per week throughout. I expect my college exams to be over before the coding period begins on 7th June. I'm currently interning at a company as a Software Engineer, but I believe it won't hamper my commitment to GSoC. After the Summer of Code, I'll keep contributing with code as well as other things like docs. The community is very active and new work is always available.

Programming Experience

Being in my final year, I've accumulated a fair amount of programming experience until now. I have worked on multiple projects across different domains like mobile app development, backend web development, and machine learning.

I love participating in hackathons and developing solutions. Our team recently won Smart India Hackathon (India's biggest hackathon), where I was the team leader. We worked on a Machine Learning based solution for calculating a safe speed for a vehicle to drive at, based on its surroundings. We're currently writing a paper on our approach and hope to get it published. We also went on to win the ASEAN India International Hackathon a few weeks ago.

My experience with Python started with learning backend web development using Django. I've gotten proficient with it now. I'm currently interning at a company where I'm the primary person tasked with writing REST APIs with Python. A few months ago, we also developed the initial stage of a mutual fund portfolio management portal, whose development was sponsored by Principal Global Services, a Fortune 500 FinTech company. The entire backend for this application was written using Python.

I have some more Python experience related to machine learning, much of it from my previous internship at Resolute AI in Bangalore, India. This mostly entailed work with libraries like TensorFlow, Keras, numpy, pandas, etc. I worked on developing a model that tracked PPE kit compliance for people working in the healthcare industry.

I'm involved in the clubs at my campus like the Google Developer Student Club as the Technical Head. We organize and conduct technical workshops for fellow juniors. Through this, I've had the opportunity to reach hundreds of students and help them through their journey.

Some of my Python projects:

  1. A desktop GUI app for Object Recognition https://github.com/adwaitbhope/object_recognition_poc

  2. A client-side app for running Object Detection https://github.com/adwaitbhope/sih-client-side-object-detection

  3. REST backend for a mobile app https://github.com/adwaitbhope/denizen-backend

Here is a summary of some relevant subjects I studied at college:

  1. Engineering Mathematics (I, II and III), Engineering Physics
  2. Data Structures & Algorithms, Object-Oriented Programming
  3. Computer Graphics, Engineering Graphics
  4. Design & Analysis of Algorithms, Theory of Computation
  5. Software Modelling & Design, Software Testing & Quality Assurance

These courses have strengthened my grasp of core computer science concepts and programming logic. Also, I believe that some of these courses like Mathematics, Physics, and Graphics will specifically help me with the science side of things as I contribute to this organization.

Also, here is a link to my resume for reference.

Contributions to sunpy

  1. (Merged) Fixed a bug in DataManager https://github.com/sunpy/sunpy/pull/5089

  2. Added namespaces for downloaded files https://github.com/sunpy/sunpy/pull/5111

  3. (Misc.) Invalid link on the GSoC webpage of this project https://github.com/OpenAstronomy/openastronomy.github.io/issues/285

  4. (Misc.) Fixed a typo in a function doc reference https://github.com/sunpy/ndcube/pull/419

Why This Project

I'm currently working on my final year academic project with our mentor from Persistent Systems Ltd., India. We are collecting spectral data of galaxies in the optical region from the Sloan Digital Sky Survey, and working on setting up a Machine Learning pipeline that can predict the age of the galaxy. In fact, we are using astropy, one of the sub-organizations here, to handle the FITS files.

I have always been excited about stars, galaxies, and other astronomical entities, and keep reading about them. All of the above project exposure piqued my interest, and I was enthralled when I discovered OpenAstronomy as one of the organizations.

In our project, we had to resample all the spectral data on the same grid so that it can be fed to the ML algorithms. We used the Python package pysynphot to achieve this. Upon seeing a similar project under OpenAstronomy, it was a no-brainer for me. I understand why this functionality would be helpful since I had to experience it myself firsthand. I find it exciting that this will make it convenient for scientific researchers to tinker with their data.

  • Why are you suited to work on this project? As stated above, I understand the motive behind this project, I have good experience with Python and programming in general, and I also have some exposure to Astronomy. I have made contributions to sunpy, and I feel comfortable with the math that goes behind these WCS transformations. I believe this makes me a strong candidate for the project.

Existing Work

There are other packages (like pysynphot that I mentioned above) that can resample data, but Astropy's reproject is probably the most helpful one for ndcube. It will accept an NDCube along with a target WCS and will project the data on it accordingly. This will make it very easy for ndcube to support resampling. The challenge to construct a WCS object with the necessary modifications remains, as reproject will not help with that.

Currently, there are multiple algorithms supported by reproject like Interpolation, HEALPIX Projection, DeForest Adaptive Resampling, etc. For Interpolation, it supports "nearest-neighbour", "bilinear", "biquadratic", and "bicubic".

What I've Explored

I studied a bit about different coordinate systems and WCS from this article that @nabobalis shared with me. It provided me with a clearer understanding of how these transformations work. I have put some code together, trying to resample an NDCube using bicubic interpolation. I upscaled the wavelength axis by a factor of 10, manually creating a new WCS that supported this transformation. For this, I modified the CDELT parameter that corresponds to pixel width (for real-world values). I have also plotted a slice of the cube to compare the result.

The code and the graph can be found at this gist: https://gist.github.com/adwaitbhope/d056fd5a5a8eb5781a9ccc5615a644f0.

  • How do you plan to implement the project? First, I plan to closely study different use cases for resampling. It is key to identify the necessary type of WCS transformations. I will be studying more about this to figure out how to create a new WCS that supports the resampled data. My first step would be to get an API up and running, that solves a basic use case, something like a Minimum Viable Product. This can involve modifying NDCube's other parameters like extra_coords. The API can then be extended to more complicated use cases. There are some more things to plan before starting with the implementation, such as deciding whether to support only integer values as the scaling factor. This will largely depend on how we are able to use reproject with ndcube.

  • Does the project include API changes? Yes, this project will introduce a new API. In the original issue that referenced this topic, @DanRyanIrish (one of the mentors) has ideated a possible API. Under the hood, ndcube will construct a new WCS object and resample the data along with other parameters like extra_coords. This will make a powerful API that will let users manipulate all their data with a single function call. However, I don't suppose that there will be any breaking change to the existing API.

  • Will you need additional software package requirements? Yes, the primary requirement will be Astropy's reproject, which is released as a separate package here. This package implements the necessary mathematical algorithms we can use for resampling, thereby reducing effort and redundancy. Its API is easy to use with NDCube, as demonstrated in the gist that I have linked above.

Timeline

Period Plan Date
Community Bonding Period
  • I'll utilize this time to keep contributing to sunpy and ndcube in other issues or PRs. This will help me get familiar with the community and understand how things work.
  • If possible, I'll try to get a headstart on understanding the math behind WCS transformations.
 17th May - 7th June
Coding Period 7th June - 16th August
Week 1
  • Study different coordinate systems and equations that enable WCS to map pixel coordinates to/from real-world coordinates.
 7th June - 14th June
Week 2
  • Study the codebase of ndcube and understand how astropy.wcs is coupled with NDCube for operations like slicing, visualizing, etc.
  • Understand how extra_coords and global_coords behave.
 14th June - 21st June
Week 3, 4
  • Discuss and formulate a basic use case and write code that solves it.
  • Try to visualize the difference by plotting the data.
  • Create a PR that introduces this new API, along with necessary documentation. Try to get feedback from the community.
 21st June - 5th July
Week 5
  • Formulate more use cases and try out some code to understand what needs to be done.
  • Figure out what WCS transformations will be necessary in such cases.
 5th July - 12th July
First Evaluation 12th July - 16th July
Week 6, 7
  • Write code for the above use cases and write basic documentation and tests.
  • Create a PR with this functionality. Get feedback from the community.
 12th July - 26th July
Week 8
  • Make necessary modifications to the API based on the received feedback.
 26th July - 2nd August
Week 9
  • Improve documentation and tests. Create PRs for both.
 2nd August - 9th August
Week 10
  • Buffer time to fix things in case something comes up during the tests.
  • Merge leftover code and facilitate the release of a new version of ndcube with complete resampling support!
 9th August - 16th August
Submission and Final Evaluation 16th August - 23rd August
Post Google Summer of Code
  • Keep contributing to sunpy and ndcube. Maintain the resampling functionality.
  • Enhance it if necessary by adding more algorithms, perhaps by contributing to astropy's reproject.

What I Wish To Gain From Google Summer of Code

Primarily, I aim to gain more experience working with open-source communities. I have contributed a bit to other repositories before, and hope that Summer of Code serves as an entry point for me to give back more to the community.

Secondly, I'm excited about being mentored during this project. Even during my recent contributions to sunpy, I understood that developing software at this level is different from academic projects. I want to grow these skills and hope to learn the difference between good, bad, and great code.

Add a custom footer

SunPy

Clone this wiki locally