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

what's the difference between the same game with different version : v0,...,v4 #552

Closed
initial-h opened this issue Aug 30, 2018 · 5 comments
Labels

Comments

@initial-h
Copy link

i find there are many versions of the same atari game,like PongNoFrameskip-v4 and so on.

what's the defference and where i can find the introduction?

@pzhokhov
Copy link
Collaborator

pzhokhov commented Sep 12, 2018

Very good question (although more gym than baselines), had it myself more than once :) @christopherhesse @gdb @joschu who'd be the best person to route this to?

@christopherhesse
Copy link
Contributor

I don't know the answer to this, but there are some notes in the gym changelog for a couple of the v* suffixes: https://github.com/openai/gym#whats-new

@gdb
Copy link
Collaborator

gdb commented Sep 12, 2018

We bump the version whenever there is a semantic change to the environment, no matter how slight. That means you can always trust that PongNoFrameskip-v3 has exactly the same semantics as the version used by some paper. That being said, you should consider all older versions to be deprecated and just use the most recent one — that is, you should think of older versions like a history that is accessible but not intended for normal use. Accordingly, we don't always document the difference between the versions, as you should always be using the most recent one except when reproing a historical result.

@KristianHolsheimer
Copy link

For Pong there's a nice video that makes the difference immediately clear: https://www.youtube.com/watch?v=zp72VTr730s

@rfali
Copy link

rfali commented Dec 22, 2021

Breakout-v0 vs Breakout-v4
v0: repeat_action_probability of 0.25 (meaning 25% of the time the previous action will be used instead of the new action)
v4: repeat_action_probability of 0 (always follow issued action)

Breakout-v4 vs BreakoutDeterministic-v4 vs BreakoutNoFrameskip-v4
game-vX: frameskip is sampled from (2,5), meaning either 2, 3 or 4 frames are skipped [low: inclusive, high: exclusive]
game-Deterministic-vX: a fixed frame skip of 4
game-NoFrameskip-vX: with no frame skip

Breakout-v4 vs Breakout-ram-v4
game-vX: Observation Space (128,). Observation is given as a vector of size 128.
game-ram-vX: Observation Space (210,160,3). Observation is given as pixels of Width:210, Height:160, Depth:3

The atari environment source code has been removed from Gym [AFAIK] and you can see it on the ALE's GitHub. See source code here

Other useful references:
This video
This GitHub Issue
ALE v0.7 Blog

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

6 participants