Notebooks

We use jupyter notebooks for debugging, finding hyperparameters, and investigating results. Specifically, we created the following notebooks:

debug_environments.ipynb:

Manually walk through the environments, experiment with configurations, and debug.

debug_wrapper.ipynb:

Step-wise observe the observations of the different wrappers for debugging.

preprocess.ipynb:

Find the hyperparameter configuration with the best reward averaged over multiple seeds and combine multiple results files.

training.ipynb:

Visualize the training progress and outcomes.

evaluation.ipynb:

Visualize the results of the evaluation rollouts.