Community Bonding: Week 3
Preliminaries
I identified some errors while using the existing frameworks. This week, we will continue to fix the issues found and discussed during the meeting. I also verified the inference time for PilotNet model (PyTorch implementation), the difference between the CPU vs GPU execution is (0.0265 vs 0.0192) seconds (mean inference time) for simulation of 1 minute.
When we save stats of a simulation a window shows all metrics in details. We will also use this in later phases to evaluate and compare our optimized models. I describe our new objectives next.
Objectives
- Solve the issue Error while trying to save stats with DL-torch.yml config #368 and submit a PR
- Fix the issue Not utilizing GPU when running simulation #369.
- Update PR updated package versions for python3.10 #46 to remove additional packages and include installation instructions in REAMDE
- Update PR Use new dataset #49 - Add feature to let user specify each dataset directory and normalize label values.
- Update PilotNet brain in BehaviorNet to unnormalize (expand) the predicted values
- Implement evaluation, on a validation set, in the training scripts of models
Issues and Pull requests.
- Submitted PR Removed: unnecessary exp details #370 to fix issue #368.
- Solved issue #369 by PR Update: brain file to use gpu #371.
- Updated PR updated package versions for python3.10 #46.
- Updated PR Use new dataset #49.
- Create a PR unnormalized prediction value #372 to unnormalize (expand) the predicted values of PilotNet brain in BehaviorNet.
- Updated PilotNet training script with PR Adding validation set for model selection #50.
The execution
This week was dedicated to resolves the previously found issues. Moreover I also included new features which would be helpful in conducting experiments and evaluation. For the training part, I worked on finalizing installation packages, enable training scripts to use GPU, also update it to use new training datasets and evaluate on a validation set. Furthermore, the inference scripts on BehaviorMetrics were updated accordingly.
References
[1] https://github.com/JdeRobot/BehaviorMetrics
[2] https://github.com/JdeRobot/DeepLearningStudio