Help Solving Version-Spaces(AI) Problem

Question

I am new in AI(Artificial Intelligence) learning Version spaces and i need help in solving sum tasks. Am using a software which will do the learning process so my task is to learn and understand what is going on(i mean why the software produces such results) and etc..

My task is i have a brand of cars:

Brand   Likes
_____   _____
opel    yes
toyota  no
bmw     yes
ford    yes
nissan  no

and my questions are:

1)how can i learn which car brand a person likes? my understanding is the first brand(opel) in the examples should be positive(yes) or?

2) how can i learn which car brand a person dislikes?
should i make the first brand negative(no)?

3)
how can i create two hierarchies so that they can be used to learn which car brands a person likes and which ones he dislikes?

UPDATE I need help for the following requirements also:

1)
Make up two different hierarchies for learning the examples that both hierarchies allow you to learn which brand the person likes, while learning which brands the person dislikes can be learned with only one hierarchy and cannot be learned with the other hierarchy.

2)
What can you be concluded about when it is possible to learn the opposite concept and when not?

---

Please am learning how the thing works so be patient with me. thanks

thanks for your help.

Answer 1

Version spaces were a kind of academic concept that were never really useful. Some disadvantages are:

1. Limited representation -- basically either 1-CNF (single conjunct) or 1-DNF (single disjunct)

2. Can't handle noise

3. Exponential time and space.

For 1-CNF descriptions, there is the simple "take the intersection" algorithm (which works exactly as you would imagin, at least in a 2-D or 3-D space) which is a pac learning (probably approximate correct) algorithm that is polynomial in the number of examples.

However, the limited representation is what really kills it -- even the "take the intersection" algorithm isn't used (also known as maximally specific conjunctive generalization or MSCG) because of the limited representation.

Instead, things like decision trees, decision lists, neural nets, genetic algorithms SVG algorithm are in common use today.