Dataset loading

Goal: One sentence

What I did:
- Got torch dataset and dataloader working with hdf5 data
- transformed data into resnet format
- Finished first iteration of model design
- Wrote training and testing loop
- Trained first model and visualized losses

What worked:


- Starting to feel familiar with python matrix manipulation
- Model appears to converge! Unsure how good performance will be on actual tasks
- model has 0 temproal context, so i predict that it ill be very jittery, especially when there are no green cubes in sight

What failed:
- Data loading is main bottleneck with training
- 95% of training cycle time is waiting for dataloader
- model stopped improving test and training loss after ~epoch 2
- maybe this is fine?

Key learning:
- Data loading can significantly slow down training

Next session:
- Test model in simulation
- need to make a new controller class that inferences model

Time spent: 5 hours