search

BerkeleyAutomation / fog_x

Apache License 2.0
9 stars 4 forks source link

Preferred way of loading RT-X files with a local buffer #14

Closed utn-blei closed 2 months ago

utn-blei commented 3 months ago

Hello,

thanks for the good work!

What is the preferred way of loading the RT-X dataset with a local buffer? When using the _load_rtxepisodes function, the dataset is freshly downloaded from the google gs server, whenever we run the code. This slows down training significantly.

I could overwrite the dataset2path function, however this is probably not the intended way of using the framework :) We're happy to contribute if this involves implementing a new feature.

Greetings,

Yannik

KeplerC commented 3 months ago

Hi Yannik,

Thanks for your interest! I'm wondering #15 would solve your issue.

I have already merged with main branch, so you should be able to clone and install from source. I may plan for a pip package release by the end of the week.

Feel free to reopen the issue

utn-blei commented 3 months ago

This helps, thanks for the quick reply! :)

utn-blei commented 3 months ago

Looks like we introduced an error here ... when running

dataset.load_rtx_episodes( name="berkeley_autolab_ur5", additional_metadata={"collector": "User 2"} )

python yields

TypeError: Dataset._build_rtx_episodes_from_tfds_builder() got an unexpected keyword argument 'split'

The solution should be as easy as adding split: str = "all", in line 423, dataset.py

Could you verify?

utn-blei commented 3 months ago

PS: I cannot reopen the issue - seems like it is disabled for this repo

NotAFood commented 2 months ago

Hi @utn-blei, I've made you an outside collaborator on this project so that you can reopen your own issues with it in the future. Unfortunately this has to be done on a person-by-person and repo-by-repo basis but should solve your problem for now.

~ AUTOLAB Systems Administration

utn-blei commented 2 months ago

Hi @NotAFood, thanks a lot! @juelg and I would be happy to contribute in the future. Is there any specific way you handle the discussion of new features? Greetings from UTN in Germany :-)

NotAFood commented 2 months ago

@utn-blei , I've activated the discussions tab for the repository if you would like to discuss new features. Any post on there should notify the other repository members.