Markov decision problems and Dynamic programming

Question 1

A transition has ___

Answer 1

The current state, current action, reward received and next state

Question 2

We shouwld always leave all the possible and impossible actions for every state?

Answer 2

Yes, and if the action is impossible to do then the output will be doing nothing

Question 3

If we know the present then the future is ___ of the past. This means that we don’t need the past to decide what will happen next

Answer 3

independent

Question 4

To describe our problem we need ___

Answer 4

State space
Action Space
Reward function
Transition Probabilities

Question 5

Its better to design the reward in a ___

Answer 5

abstract way

Question 6

In practice when the agent is learning, we showld give ___ in the reward to serve as a ___ to the agent

Answer 6

hints

guide

Question 7

Optimality criterion defines the criterion to select between ___

Answer 7

actions

Question 8

Discount factor acts as an ___ rate

It makes the agent prefer rewards ___ than ___

Answer 8

inflation
sooner
latter

Question 9

Policy is a mapping that maps everything the agent as ___ (___) to distributions over ___

Answer 9

seen so far
history
actions