Spaces are a way to represent the environment it's action and observation space. An observation can for example be a discrete observation in the range of numbers between 0 and 1, but it could also be an image that was captured.
https://github.com/openai/gym/blob/master/gym/spaces/discrete.py
A discrete space will allow you to represent a linear space (e.g. 0, 1, 2, ..., n - 1 for a given n), this can be used in situations where you would need to represent 4 actions, which would be written as Discrete(4).
https://github.com/openai/gym/blob/master/gym/spaces/box.py
As defined by OpenAI, a box represents the Cartesian product of
A square would thus be represented as Box(2, ) where we would observe 2 values (e.g. {x, y} values). Similar, Box(3, ) would represent 3 values (e.g. {x, y, z}).
In the definition of our box, we include a parameter to define the boundaries of the dimensions.
For optimality, we include a Box implementation for Float and Int datatypes.
https://github.com/openai/gym/blob/master/gym/spaces/tuple.py
The Tuple space is a combination of other spaces, this allows you to represent more complex states that exist out of multiple substate.
Example: Let's say that we want to represent the following state:
- Image (64x64 px RGB)
- {x, y, z} coordinates in a cube between -50 and 50
- { Dead, Alive, Resurrecting } represented as { 0, 1, 2 }
We could write this as: Tuple([ Box(0, 255, shape=(64, 64, 3)), Box(-50, 50, shape=(3, )) ])