change create dataset from buffer to ensure same behavior with create data from env #146
Conversation
|
Thanks for this contribution, and I agree with the direction. It looks correct to me. Please write a test which makes sure a partial episode is correctly marked with a truncation when calling |
|
Hi @balisujohn , |
|
Once typos I pointed out are fixed and pre-commit is passing, this is ready to merge :^) |
|
Hi, finally I turned back to fix the problem. Could you check it again? |
…fixbug into fix_create_from_buffer
There was a problem hiding this comment.
@im-Kitsch thanks for doing this; it seems good. However, would you mind to wait until #133 is merged?
The problem will be solved by that PR, but we may want to add your test
| else: | ||
| assert ( | ||
| eps_buff["terminations"][-1] or eps_buff["truncations"][-1] | ||
| ), "Each episode must be terminated or truncated before adding it to a Minari dataset" |
There was a problem hiding this comment.
it looks like this branch is not very useful, I would remove it
@younik Hi, thanks for the reply, looks after #133 , merging data would diable |
|
Right @im-Kitsch, thanks for your help :) |
Description
Hi,
I think it would be better to keep
create_from_bufferandcreate_from_collevtor_envhaving same bahavior. If we create dataset from collector env, the last episode will be marked as truncated if it's incomplete trajectory. But if create dataset from replay buffers, incomplete trajector will be rejected.Type of change
Screenshots
Checklist:
pre-commitchecks withpre-commit run --all-files(seeCONTRIBUTING.mdinstructions to set it up)pytest -vand no errors are present.pytest -vhas generated that are related to my code to the best of my knowledge.