search

openai / large-scale-curiosity

Code for the paper "Large-Scale Study of Curiosity-Driven Learning"
https://arxiv.org/abs/1808.04355
805 stars 180 forks source link

Does normalized rewards works with other Agents for Attari ? #13

Open zafarmah92 opened 5 years ago

zafarmah92 commented 5 years ago

Using the normalized reward (#6 ) with the other agent's, taking the example of A2C where the discounted rewards are used on the extrinsic reward.

  1. Now to which extent we need to normalize the intrinsic rewards as the forward loss (#3 ) tend to get low with the passage of time.
  2. Second, Does the scaling of rewards also requires advantage normalization. ? In the perspective of agents A2C, ACKTR etc