Reinforcement learning experiments

Instructions for reproducing the RL results in the paper.

Reproducing figure

See figs/ for details on reproducing the RL figure in the ICLR paper.

Run a single experiment (e.g SFR) for a single random seed with:

python train.py +experiment=sfr-sample ++random_seed=100

You can display the base config using:

python train.py --cfg=job

and an experiment's config with:

python train.py +experiment=sfr-sample --cfg=job

Run all of the RL experiments for 5 random seeds with (you'll need a cluster for this):

python train.py -m +experiment=sfr-sample,laplace-sample,ensemble-sample,ddpg,mlp ++random_seed=100,69,50,666,54