Each experiment uses 3 seeds. The parameters used for DQN are the same parameters as described in the original paper, except for the optimizer (changed to ADAM) and learning rate (1e-4) used.

Breakout DQN - single worker

python3 coach.py -p Atari_DQN -lvl breakout

Pong DQN - single worker

python3 coach.py -p Atari_DQN -lvl pong

Space Invaders DQN - single worker

python3 coach.py -p Atari_DQN -lvl space_invaders