SPaCE

Experiments for the BIG@ICML 2020 workshop paper "Towards Self-Paced Context Evaluations for Contextual Reinforcement Learning"

@inproceedings{eimer-bigicml20,
  author    = {T. Eimer and A. Biedenkapp and F. Hutter and M. Lindauer},
  title     = {Towards Self-Paced Context Evaluations for Contextual Reinforcement Learning},
  booktitle = {Workshop on Inductive Biases, Invariances and Generalization in {RL} ({BIG@ICML}'20)},
  year = {2020},
  month     = jul,
}

Setup & Usage

To run the experiments, you need to install the dependencies:

pip install -r requirements

The included notebooks for plotting and generating new instance also require jupyter to be installed:

pip install jupyter

To train a SPaCE agent on our provided instances, you can run the following on a GPU (does not currently work on CPU):

python src/ray_spl.py --mode spl --instances features/cpm_train.csv --test features/cpm_test.csv

Replacing "spl" with "rr" will result in a round robin trained agent for comparison.

Alternatively, you can use our provided slurm script.

Results included in the workshop paper

Our own training results are included in this repository. To plot the data, you can use the provided jupyter notebook.