// Parameter for gym.make()
// the name must be found in the gym registry
// Parameter for <env.wrapper.FrameStack>:
// - "concat": apply np.concatenate to frames, e.g. concatenate 4 images with shape (1, 84, 84)
// - "stack": apply np.stack to frames (a new stacking dimension will be created)
// - null: no frame preprocessing will be applied
// Parameter for <env.wrapper.FrameStack>:
// - {int}: the number of successive frames used for frame_op; 4 is common.
// - null: when frame_op is null
"frame_op_len": int|null,
// Parameter for <env.wrapper.NormalizeStateEnv>
// apply on-line normalization to state
// Parameter for <env.wrapper.ScaleRewardEnv>
// - "sign": transform reward wiht np.sign to yield -1, 0, or +1
// - {int|float}: scale reward by the given number
// - null, no reward scaling is applied
"reward_scale": null|str|int|float,
// Parameter for <env.vec_env.ShmemVecEnv>
// If > 1, this will apply asynchronous parallelization to spawn vector environments
// The maximum time steps per episode
// - {int|float}: use the given number as the maximum time steps
// - null: default to the maximum time steps defined in the environment
// The maximum frames (total time steps) to run a sess