Applying deep Q learning to Atari games