This report considers various negative sampling techniques.

Note that DSSMs which provide the IL reward were trained on datasets containing negatives for all states. Therefore following experiment serve more as a proof of concept. Moreover, only MountainCar and Acrobot are considered as agent learns to solve CartPole as long as reward as positive.

This report considers various values of temperature and learning rate for LogSoftMax Reward.

Agents trained on CartPole with different rewards.

Size of Experience Replay is 10. Fixed subset of size 10 is chosen from the expert dataset.