ZiwenZhuang / rllab-learn2learn

rllab-learn2learn

This repo is forked from https://github.com/florensacc/rllab-curriculum which possibly from rllab.

Installation

We have only installed it through anaconda, which is the only way we could provide. Unfortunately, the initial rllab is deprecated. And for the safety and easy to locate problem while installing, please install mujoco-py manually in the virtual environment.

cd path/to/rllab-learn2learn
conda env create -f environment.yml

# manually install mujoco-py in the virtual environment
# manually upgrade tensorflow to meet the dependency requirement of tensorflow-probability

Error & Solution

• If you cannot use GPU version of tensorflow, please change tensorflow-gpu to tensorlfow in environment.yml
• If you are going to use mujoco environment, which is commercial, you need to acquire a key and install it. (please notice the mujoco version needed by mujoco-py)
• If you see file not found error with patchelf prompted, please install it manually with sudo apt install patchelf
• If you get error on incompatible ipython dependency, please manually upgrade ipython via pip. (in the virtual environment)
• If you see other dependencies that are not solved by conda itself, just install the version that meets the requirement. (there should not be too much)

You can read more to setup rllab and see documentation at https://rllab.readthedocs.org/en/latest/.

Changes

All experiments code are at learn2learn directory. Other from that, they are inherited from upstream.