GSoC 2025: Discussion on "Lazy Trajectory Analysis with Dask and a Lazy Timeseries API"

[Abdulrahman-PROG]

Hi @ljwoods2, @orbeckst, @yuxuanzhuang, and the MDAnalysis team,

I’m Abdulrahman Yasser Elbanna , a Bachelor’s student in Artificial Intelligence at Kafrelsheikh University, Egypt. I’m very interested in contributing to the "Lazy Trajectory Analysis with Dask and a Lazy Timeseries API" project for GSoC 2025. I’ve already contributed to MDAnalysis through PR #4951 ( for issue #3743), which is currently under review.Additionally, I contributed to MolecularNodes by updating the van der Waals radii for elements 47–103 in PR link, which was successfully merged. I’m excited about improving MDAnalysis’s performance in HPC environments through this project.

I have strong experience in Python, data analysis, and object-oriented programming, having worked on projects like sign language classification using CNNs and Automatic Number Plate Recognition (ANPR) with OpenCV and EasyOCR. These projects honed my skills in handling large datasets and using NumPy-like interfaces, which are directly applicable to MDAnalysis’s trajectory analysis needs. I’ve been researching the project and related issues (#4713, #4598, #4561, #2865). I understand that the goal is to implement a lazy_timeseries API using Dask for trajectory readers, provide a sample H5MD implementation, and develop a lazy analysis base with an example algorithm like RMSD. I’m currently learning Dask and exploring the existing timeseries API and H5MDReader in package/MDAnalysis/coordinates/H5MD.py.

I’ve submitted a pre-proposal for this project, where I outlined my plan to develop a Dask-based lazy reader, create a lazy_timeseries interface compatible with H5MDReader, implement a lazy RMSD analysis, achieve 90% test coverage, and reduce analysis time by 20% for large trajectories in HPC. I’d love to discuss a few key aspects of the project to ensure my final proposal aligns with your expectations:

1. Scope of lazy_timeseries API: What specific features or performance benchmarks are you looking for in the lazy_timeseries interface? For example, should it prioritize memory efficiency, speed, or compatibility with existing readers?
2. H5MD Implementation: Are there specific challenges or requirements for the H5MD implementation (e.g., handling large datasets, specific metadata support)?
3. Analysis Base and RMSD: For the lazy analysis base, would you prefer a specific design pattern (e.g., a base class with modular methods)? Also, are there particular RMSD implementations you’d like to see as a proof of concept?
4. Testing and Validation: What are your expectations for testing the lazy implementation? Should I focus on specific HPC environments or datasets for validation?
5. Proposal Focus: Are there any areas of the project you’d like me to emphasize in my GSoC proposal? For example, should I prioritize certain deliverables over others?

I’m eager to contribute to MDAnalysis and make this project a success. Any guidance or feedback would be greatly appreciated as I prepare my final proposal, due on March 24, 2025. Thank you for your time!

Best regards,
Abdulrahman Yasser Elbanna

[orbeckst]

Thanks for submitting your preproposal!