Env from text file about minigrid HOT 4 CLOSED

I agree that if there is a text-based format, it should mirror the format from the __str__ method. We may want to change the symbols used to represent the agent though, as they are non-ASCII and hard to type on a keyboard.

For creating envs with parameters, the way to do this is to import them directly, eg: from minigrid.envs import TextFileEnv(file_name).

I'm going to ask the obvious question though: do you personally have a need for this feature? Would you make use of it. Because it's not that hard to create environments just by writing code, so unless you actually have a legitimate use for this, I would say YAGNI?

from minigrid.

I agree that if there is a text-based format, it should mirror the format from the __str__ method. We may want to change the symbols used to represent the agent though, as they are non-ASCII and hard to type on a keyboard.

For creating envs with parameters, the way to do this is to import them directly, eg: from minigrid.envs import TextFileEnv(file_name).

I'm going to ask the obvious question though: do you personally have a need for this feature? Would you make use of it. Because it's not that hard to create environments just by writing code, so unless you actually have a legitimate use for this, I would say YAGNI?

Thanks it seems like we do need this feature.

from minigrid.

One thing we should do for this feature is have some round-trip testing. Grid to text, then parse the text back into a grid, and make sure the two versions are identical. I would implement all of the __str__ and parsing functionality into the grid class. That will also require us to treat agents as world objects, in order to parse the agent position as well.

from minigrid.

Closing for now.

from minigrid.