The last post

This is my last post regarding all the work I have done and my experience with RoboComp.

Summary of all the work done

I started out by increasing the speed of graph data generation for training the GNNs, after that was done I seperated the GNNs and CNNs code for cleaner and more structured code base. I provided better support for both DGL and PG frameworks in the training loop. I helped in creating a GUI toolkit for collecting the robotic dataset along with that I helped in providing more GNN architectures as a usecase for the data collected from the toolkit.

What I learned

GSoC’20 helped me a lot in learning things like:

1. Using PyTorch to write GNN architectures and ML models in general, I learned about PG and DGL.
2. How to write cleaner code so that everyone in the team can read it and continue help building it.
3. pub/sub communication between two processes.
4. Unix signal handling
5. PyQT and PySide to make GUIs.
6. How to collaborate with a team using git.

Final Work

SNGNN: Github Repository

• Pull Request : Link

Future work

A thing that can be added is a second use-case to label interactions by building on top of our toolkit.

I am thankful to my mentors Ronit Jorvekar, Pilar Bachiller, Daniel Rodriguez Criado for helping me when I was stuck. A special thanks to Luis J. Manso for helping me out in my pre-GSoC period and during GSoC.

Rishabh