EasyDeepLearn

PPO implementation details that actually matter.

hard

Answer

  • Engstrom et al. 2020 'Implementation Matters': (1) value function clipping, (2) reward scaling by running SD, (3) orthogonal init, (4) Adam with fixed lr (β2  =  0.999)({\beta}_{2}\; = \;0.999), (5) global gradient clip to 0.5, (6) GAE with λ = 0.95, (7) normalize advantages per-batch, (8) LR annealing, (9) N epochs per rollout (~4-10).
  • Many published gains are due to these tricks, not the core algorithm.
  • Always study a modern reference implementation.
Check yourself — multiple choice
  • Random
  • Value clipping, reward scaling, orthogonal init, gradient clip 0.5, GAE λ=0.95, advantage normalize, LR anneal, 4-10 epochs/rollout — Engstrom 2020 showed these matter
  • Not important
  • Just clip

PPO tricks: value clip / reward scale / orthogonal init / clip 0.5 / GAE 0.95 / adv norm / anneal / N epochs.

#policy-methods#engineering

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions