Each experiment uses 3 seeds and is trained for 10k environment steps. The parameters used for Dueling DDQN are the same parameters as described in the original paper.
coach -p Atari_Dueling_DDQN -lvl pong
coach -p Atari_Dueling_DDQN -lvl breakout
coach -p Atari_Dueling_DDQN -lvl space_invaders