Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Mimic the details of prioritized experience replay #301

Merged
merged 3 commits into from
Aug 31, 2018

Conversation

muupan
Copy link
Member

@muupan muupan commented Aug 31, 2018

This PR adds changes to mimic the details of the original paper and their implementation.

  • Errors are clipped to [error_min, error_max]. The default is [0, 1].
  • eps is added to clipped errors before applying alpha.
  • eps's default value is set to 0.01. I confirmed that they use 0.01 by asking the first author via email.

I will add changes in examples to mimic their training settings later.

epsilon is added to (absoulte) errors, not priorities
default epsilon is 0.01
TD errors are clipped by [-1, 1]
Copy link
Member

@toslunar toslunar left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@toslunar toslunar merged commit eb18687 into chainer:master Aug 31, 2018
@toslunar toslunar added this to the v0.5 milestone Sep 7, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants