We release the experiment data to facilitate future research. Each file is a dictionary keyed by (task_name/domain_name/all, method_name) tuples. Each entry contains an numpy array of shape (30, 12) (except for RLPD where the shape is (10, 12)). The array stores the success rate at a regular interval of 50K training steps for 12 seeds.
See sanity-check.ipynb for some quick examples for visualizing and plotting our experiment data.