Open llan-ml opened 6 years ago
I am curious for the use cases for this.
Right now we have found using a temporary file for XML is both useful for debugging and fast enough for us.
@machinaut During the training of RL algorithms, we may want to change the environment dynamically. Take a maze for example, we need to generate a new maze randomly in each iteration or every several iterations.
Currently, the solution is to first open a template xml file, modify the configuration, and write it into the disk. However, after a lot of iterations, the system will raise an error that a single process cannot open more files.
I think a pull request adding VFS support would be welcome.
do you have an example file which reproduces the file error? I have not yet seen it (and we also have many machines which make many environments) but maybe the error would be more informative.
Currently, we have to load model from xml file on the disk. Any plan to support load model from xml string in the memory, like https://github.com/deepmind/dm_control/blob/master/dm_control/mujoco/wrapper/core.py#L228?