Google

A multiset is a collection where elements can be repeated. This concept can be applied to represent a group of state-action pairs from reinforcement learning. For instance, a multiset containing several identical `(s, a)` pairs would be written in set-like notation as `{(s, a), ..., (s, a)}`.

Example of a Multiset of State-Action Pairs

An agent's experience is recorded as a sequence of four state-action pairs: first `(s1, a1)`, then `(s1, a2)`, then `(s2, a1)`, and finally `(s1, a1)`. How would this collection of four interactions be represented as a multiset?

An agent's experience over five time steps is recorded as the following multiset of state-action pairs: `D = {(s1, a1), (s2, a2), (s1, a1), (s3, a1), (s1, a1)}`. If this data were instead stored as a standard set, which only contains unique elements, what crucial information would be lost?

An agent's experience in an environment is collected as the following multiset of state-action pairs: `D = {(s_A, a_1), (s_B, a_2), (s_A, a_1), (s_C, a_1), (s_A, a_1), (s_B, a_2)}`. Based on this data, how many times did the agent take action `a_1` from state `s_A`?

Learn Before

Related