Community Bonding: Week 2

2 minute read


Through the discussion, we identify potential issues which need to be worked on before the start of the coding period. We also took a diver dive into the working of Dockers to continuously support the Behavior Metric tool. While I explore the Quick Start section, I identify interesting scripts and other features to visualize the performance and save the stats of the deep learning model. Furthermore, the mentors introduced me to the development of new dataset on DeepLearningStudio [2]. After I mention an improvement in the training script, we agreed to include a validation set. In the coming weeks, we would also like to benchmark the existing models on my local machine to set a baseline for optimized models.


  • Resolve dependency issue on DeepLearningStudio’s virtual environment installation
  • Fix Dockerfiles to build the workflow in Behavior Metric
  • Explore the script and report the relevance of it for the project
  • Update the PilotNet (PyTorch) model to use new dataset in DeepLearningStudio [2]

Issues and Pull requests.

The execution

My initial task includes upgrading Python version and related packages in the DeepLearningStudio repository and submitted PR updated package versions for python3.10 #46. Next, I worked on the issues with docker images in BehaviorMetric repository, by locally buidling, resolving errors and creating containers and then submitted PR Fixes failing build of Docker images (with GPU support) in the workflow #365. Building and fixing Dockers is especially time consuming (each build trial could takes hours). I encountered additional errors while using BehaviorMetrics for which I created corresponding issues and some solutions. The new dataset is relatively large (~ 11 GB), so extracting and managing will take time. I also updated corresponding scripts for PilotNet model in DeepLearningStudio to use new dataset.