Closed tanksha closed 2 years ago
Ah, I see, the MalmoWrapper
is not adjusted, I didn’t take that into account, sorry. I will send a patch for that.
The MalmoWrapper parse_world_state
returns [reward, observation, done] but I saw #37 fixes it. Closing this PR.
@ntoxeg, the error
Torch not found, torch truth value will not be available
Indicates that your image is using the singnet fork of atomspace. Since that PR is for opencog/rocca, I assume it is better to use opencog/atomspace as well. Likewise, perhaps ideally we'd want to have singnet/rocca integration test to use an image with singnet/atomspace, even though given that singnet/atomspace offers nothing that rocca uses at that point, it doesn't really matter (I'm frequently merging opencog/atomspace into singnet/atomspace anyway).
@ngeiswei yes, it would make sense but I need to figure out how to setup a workflow that handles this logic, as I can’t make a different workflow in a fork, maybe it’s possible to have two workflow files and use repository settings to manually choose one or the other, I don’t know yet.
@ntoxeg, for atomspace and such we addressed that by using circleci variables in the config file. See for instance https://github.com/opencog/atomspace/blob/b5b31542217ae9d44eeee29c475646b5726e60b0/.circleci/config.yml#L6. But as I said, since opencog/rocca and singnet/rocca usually mirror each other (and if not, it is short lived), I think it's acceptable if you use one image for both, preferably opencog's since singnet tends to be lagging a bit behind and its extra functionalities are not used at that point.
@ngeiswei I can’t use the fix you posted because I’m a single user on DockerHub — that solution relies on the fact that you have opencog
and singnet
organisations existing there. I’ll just change to use opencog
repositories since you’ve said it lags behind a bit. I might figure something out later but it also means I have to produce more images.
The reward and observation variables were reversed.