DANCE: Difficulty and Novelty Co-Driven Exploration Our code is based on RND (https://github.com/openai/random-network-distillation) and Openai baselines (https://github.com/openai/baselines).
The env you need install: python 3.6 CUDA 10.0 CUDNN 7.6.0 pip install gym pip install gym[atari] pip install tensorflow-gpu==1.14.0 pip install numpy openmpi pip install mpi4py
The following command should train an DANCE agent on Montezuma's Revenge
python run_atari.py --gamma_ext 0.999
To use more than one gpu/machine, use MPI (e.g. mpiexec -n 8 python run_atari.py --num_env 128 --gamma_ext 0.999 should use 1024 parallel environments to collect experience on an 8 gpu machine).