Skip to content
This repository was archived by the owner on May 6, 2021. It is now read-only.

Conversation

@findmyway
Copy link
Member

@findmyway findmyway commented Jun 24, 2020

Though the fix here seems to be trivial. But it fixes some big problems in prioritized experience buffer.

Before this, the loss of rainbow with atari pong:

image

After this fix:

image

The reason is that, in some cases, get(t::SumTree, v) might return ind, p where p=0.0. And it makes the rescaled weights decreased to nearly zero.

@findmyway findmyway merged commit ab738de into JuliaReinforcementLearning:master Jun 24, 2020
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant