7 - Reinforcement Learning

Question 1

Reinforcement Learning

Answer 1

Learn to take actions in an environemtn to maximise rewards

Question 2

State

Answer 2

Information from the environment

Question 3

Policy (pi symbol)

Answer 3

A map from state space (s) to action space (a)

Question 4

Reward Function (R) meaning

Answer 4

Maps each state (or state-action pair) to a reward number

Question 5

Value FUnction, (Q pi) meaning

Answer 5

Value of a state/state-action pair.

Total expected reward

Question 6

Q learning method

Answer 6

For each episode:
- Select a random initial state

While (not goal):
- Select one action for the current state
- Bellman Equation
- Set the next state as the current state.

Question 7

Equation involved in Q Learning

Answer 7

Q(s, a) = R(s,a) + Gamma * Max[Q(next state, all actions)]

s - state space
a - action space
R - reward function
Gamma is a value (?)

Question 8

Reward Matrix

Answer 8

Links states with reward values.

Eg Going to room 5 might mean giving moves to room 5 a 100 reward etc

The matrix is State (rows) by Action (columns)

Question 9

Q Matrix - rows are ? columns are ? start values are ?

Answer 9

Matrix of state (rows) and actions (columns)

All values start 0