5.Decision Diagram

7/24/2019 5.Decision Diagram

1/12

1

1

Modeling Rule Systems

Overview:

Decision tables

! Literature:

!

chapter 2

2

Development of Rule based systems

! Target a decision to prototype

! Create decision tables

! Implement using XS shell (CLIPS)

! Decide on strategy (backward vs. forward chaining)


2/12

2

3

Decision Tables

! What is a decision table ?

! Table representing complete set of conditional expressions

! where expressions are mutually exclusive

! in a predefined area

4

Example decision table


3/12

3

5

Decision Tables

! Consists of three parts! Condition rows (stubs)

Lists condition relevant to decision

! Action rows (stubs) Actions that result from a given set of conditions

! Rules

Specify which actions are to be followed for a given set of conditions

6

What kinds of knowledge ?

! Regulations, legislation,

! Business rules, corporate policy, ...

! accept/refuse orders

! discounts

!

! Expertise

!

Classification knowledge

! types of customers

! risk categories

!


4/12


5/12

5

9

Scenario

! A company wants to unify the way orders are handled:

! All orders of non star-clients with bad credit should berejected.

! If there is enough product in stock, orders should be accepted;otherwise order is put in waiting list

10

Decision diagram

! Which conditions (variables) influence decision?

! What are the values of those conditions?

! How to find out/calculate?

! What are the relationships between conditions?


6/12

6

11

Scenario:conditions and actions

! A company wants to unify the way ordersare handled:

! All ordersof nonstar-clientswith badcreditshould berejected.

! If there is enoughproduct in stock, ordersshould be accepted;otherwise order is put in waiting list

12

Decision diagram example

! Determine maximum number of rules! Number of rules = values(cond1)**values(condN)

! Rules = 2 * 2 * 2 = 8

Handle

order

Star-client stockCredit

(good, bad) (yes, no) (sufficient, insufficient)

(accept, reject, waiting-list)

Conditions:

Actions:


7/12

7

13

Decision table

credit

star-client

Stock

Orderhandling

8 rows = max number of different rules

14

Decision table

credit good good good good bad bad bad bad

star-client Yes Yes No No Yes Yes No No

Stock Suf Insuf Suf Insuf Suf Insuf Suf Insuf

Order

handling


8/12

8

15

Decision table


star-

client

Yes Yes No No Yes Yes No No


Orderhandling

Accept Wait Accept Wait Accept Wait Reject Reject

! Verify table with experts!

! All conditions?

! All values?

! Correct decisions?

16

Simplification

! Redundant rules?

Type of client does not matter stock does not matter


star-

client

Yes Yes No No Yes Yes No No


Orderhandling

Accept Wait Accept Wait Accept Wait Reject Reject


9/12

9

17

Decision table simplified

credit good bad bad

star-client - Yes No

Stock Suf Insuf Suf Insuf -

Orderhandling

Accept Wait Accept Wait Reject

18

From decision tables to rules

R1: IF credit = good AND stock = suf THENhandling = accept

R2: IF credit = good AND stock = insuf THENhandling = waiting-list

R3: IF credit = bad AND star-client = yes AND stock = suf THENhandling = accept

R4: IF credit = bad AND star-client = yes AND stock = insuf THENhandling = waiting-list

R5: IF credit = bad AND star-client = no THENhandling = reject

credit good bad bad



Order

handling

Accept Wait Accept Wait Reject


10/12

10

19

Rules to CLIPS

! R1: IF credit = good AND stock = suf THEN

handling = accept

(defrule credit_rule

(credit good)

(stock suf)

=>

(assert (handling accept)))

! Syntax!!

20

Combining decision tables

! Not all values are directly determined, some times decision

is needed.

! Scenario revisited:

! When is a client a star-client?

! How to calculate if stock is sufficient?

! How to know quality of credit?

!

! Ex:

Star clients have account age of more than 2 years, or have anaverage order amount higher than 5K

Stock is sufficient if requested product is availablein thestorehouse in a higher quantity than ordered


11/12

11

21

Extending decision diagram

Handle

order

Star-client stockCredit

Account age avg-amount product quantity

(good, bad)

(yes, no)

(sufficient, insufficient)

(accept, reject, waiting-list)

(yes, no) (yes, no)(=5K)(=2)

22

Combining decision tables

credit good bad



Order handling Accept Wait Accept Wait Reject

Account age >=2


12/12

12

23

Next lecture: 29 September

Frames/ontologies

Documents

5.Decision Diagram