Code for the paper "Learning Temporal Distances: Contrastive Successor Features Can Provide a Metric Structure for Decision-Making."
- Build conda environment:
conda env create -f environment.yml - Activate the environment:
conda activate metrics - Install pip dependencies:
pip install -r requirements.txt --no-deps - Export environment variables
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CONDA_PREFIX/lib
Check run.py for available tasks and specific hyperparameters. You can turn on --debug and --run_tf_eagerly to run the code in debug mode.