dniku / rl-attention

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Implement skeleton based on stable-baselines

dniku opened this issue · comments

Approximate requirements:

  1. Script that trains an agent with one of the existing algorithms in stable-baselines
  2. Runs locally, no need to adapt for cloud
  3. Fixes random seed
  4. Specifies network architecture (might be a copy of a predefined one), not imports it from the library
  5. Saves model after training
  6. Saves metrics after training (similar to ones reported in LTIAA paper)

Let's make attention visualization a part of another issue (inb4: gym has video recording).

Remains:

  • Import code to repo
  • Write instructions for launching