Transition

class Transition(s0: torch.Tensor, a: torch.Tensor = None, r: torch.Tensor = None, s1: torch.Tensor = None, done: bool = False, x0: torch.Tensor = None, x1: torch.Tensor = None, a1: torch.Tensor = None, a_dist: torch.distributions.distribution.Distribution = None, o: Option = None, ρ: float = 1.0, buffer_index: int = - 1, info: dict = {})

Bases: tuple

An experience tuple.

The main purpose of the class is to hold \((s,a,r,s')\) tuples. Additional info required by some algorithms can also be added.

TODO: consider using dataclasses from python3.7

pros:

mutability

cons:

potentially larger memory footprint incompatible with python3.6 need to write a custom __iter__ method