Decision Models Prof Juran 2 Overview Evolutionary Solver Genetic Algorithm Advertising Example Product Design Example Conjoint Analysis Decision Models Prof Juran 3 Nonlinear Problems ID: 532202
Download Presentation The PPT/PDF document "Session 5b" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Session 5bSlide2
Decision Models -- Prof. Juran
2
Overview
Evolutionary Solver
(Genetic Algorithm)
Advertising Example
Product Design Example
Conjoint AnalysisSlide3
Decision Models -- Prof. Juran
3
Nonlinear Problems
Some nonlinear problems can be formulated in a linear fashion (i.e. some network problems).
Other nonlinear functions can be solved with our basic methods (i.e. smooth, continuous functions that are concave or convex, such as portfolio variances).
However, there are many types of nonlinear problems that pose significant difficulties.Slide4
Decision Models -- Prof. Juran
4
Nonlinear Problems
The linear solution to a nonlinear (say, integer) problem may be infeasible.
The linear solution may be far away from the actual optimal solution.
Some functions have many local minima (or maxima), and Solver is not guaranteed to find the global minimum (or maximum).Slide5
Decision Models -- Prof. Juran
5
3 Solvers
Simplex LP Solver
GRG Nonlinear Solver
Evolutionary SolverSlide6
Decision Models -- Prof. Juran
6
Radio Advertising Example
Music radio WABC has commercials of the following lengths (in seconds):
15, 15, 20, 25, 30, 35, 40, 57
The commercials must be assigned to 60-second breaks. What is the fewest number of breaks that are needed to air all of the commercials?Slide7
Decision Models -- Prof. Juran
7
Managerial Problem Definition
Decision Variables
Which commercials get assigned to which programming breaks.
Objective
Minimize the total number of breaks.
Constraints
Every advertisement must be aired.
No break can be longer than 60 seconds.Slide8
Decision Models -- Prof. Juran
8
Formulation
Decision Variables
Define
x
i
to be an integer variable identifying the break to which commercial
i
is assigned. For example, if commercial 1 is assigned to break 4, then
x
1
= 4.
It should be clear that we won’t need any more than eight breaks, because there are only eight commercials. These eight
x
variables are the decision variables. Slide9
Decision Models -- Prof. Juran
9
Define
y
j
to be a binary variable, such that
y
j
= 0 if no commercials are assigned to break
j
, and
y
j
= 1 if any commercials are assigned to break
j
.
Define
v
ij
to be a binary variable such that
v
ij
= 1 if commercial
i
is assigned to break
j
, and
v
ij
= 0 otherwise.
Define
w
i to be the duration of commercial i.
FormulationSlide10
Decision Models -- Prof. Juran
10
Formulation
Note that the duration of break
j
is equal to
Let
t
j
be the amount of “overtime” in break
j
. That is,Slide11
Decision Models -- Prof. Juran
11
Formulation
Objective
Our objective, then, is to:
Minimize Z =
where
m
is a “large number”. Slide12
Decision Models -- Prof. Juran
12
Formulation
This is a good example of an advanced optimization trick: taking a constraint and building it into the objective function.
It doesn’t really matter what value we use for
m
, as long as it is sufficiently large as to prevent any
t
j
> 0. As it happens, in this problem
m
= 100 works fine.
Slide13
Decision Models -- Prof. Juran
13
Formulation
Constraints
For all
x
i
, 1≤
x
i
≤ 8.
For all
x
i
,
x
i
is an integer.
For all
y
j
,
y
j
is binary.Slide14
Decision Models -- Prof. Juran
14
Solution MethodologySlide15
Decision Models -- Prof. Juran
15
Solution Methodology
The objective function is in B25, including a penalty of 100 units per second if any breaks go over 60 seconds.
The decision variables (
x
i
) are in the range E5:E12 (in the spreadsheet shown, all commercials are assigned to break 1, so
x
i
= 1 for all
i
).
The range B5:B12 contains the durations of each commercial (
w
i
), and the range B16:B23 uses the Excel SUMIF function to calculate the duration of each commercial break.Slide16
Decision Models -- Prof. Juran
16
Solution Methodology
The range C16:C23 keeps track of which breaks have any assignments (the
y
i
variables), while the range D16:D23 keeps track of how much the breaks go over the maximum limit (the
t
j
variables). Recall that the
y
i
and
t
j
variables are the basic ingredients of the productive function.
Notice how the use of the IF function in C16:C23 precludes the need to have an explicit binary constraint in Solver for the
y
i
variables.Slide17
Decision Models -- Prof. Juran
17
Solution Methodology
The standard simplex algorithm (Solver’s default method) won’t work on this problem. The GRG Nonlinear algorithm will make an honest effort, but is likely to give up without finding the optimal solution. This is because of our use of MAX, IF, and SUMIF functions, resulting in discontinuities in our productive function and constraints as functions of the decision variables.
However, the Evolutionary Solver, a genetic algorithm, can do a good job with a problem like this.Slide18
Decision Models -- Prof. Juran
18Slide19
Decision Models -- Prof. Juran
19
Solution Methodology
The Evolutionary Solver operates in a completely different way from the other types. Instead of searching in a structured way guaranteed to reach the optimal solution, genetic algorithms operate somewhat like biological evolutionary processes, with some degree of randomness in the steps taken from one solution to the next.
In a finite period of time, the Evolutionary Solver is not guaranteed to find the optimal solution, but it will find very good solutions and try to improve upon them. Slide20
Decision Models -- Prof. Juran
20
Optimal SolutionSlide21
Decision Models -- Prof. Juran
21
Conclusions
The solution indicates that commercials 1, 2, and 5 should go in one break, 3 and 7 should go in another, 4 and 6 should go in another, and 8 should go by itself.
A reasonably bright person could solve this problem in their head, of course. The trick here was to set it up so that a computer could solve it, providing a method for the solution of much larger problems with the same basic structure.Slide22
Decision Models -- Prof. Juran
22
Product Design Example
Conjoint Analysis
is a multivariate technique used specifically to understand how respondents develop preferences for products or services.
It is based on the simple premise that consumers evaluate the value or utility of a product (real or hypothetical) by combining the utility provided by each attribute characterizing the product.
-- Prof. Pradeep Chintagunta, Univ. of ChicagoSlide23
Decision Models -- Prof. Juran
23
Conjoint Analysis
Conjoint Analysis is a
decompositional
method. Respondents provide overall evaluations of products that are presented to them as combinations of attributes. These evaluations are then used to infer the utilities of the individual attributes comprising the products.
In many situations, this is preferable to asking respondents how important certain attributes are, or to rate how well a product performs on each of a number of attributes.Slide24
Decision Models -- Prof. Juran
24
Conjoint Analysis
After determining the contribution of each attribute to the consumer’s overall evaluation, one could
Define the product with the optimal combination of features
Predict market shares of different products with different sets of features
Isolate groups of customers who place differing importances on different features
Identify marketing opportunities by exploring the market potential for feature combinations not currently available
Show the relative contributions of each attribute and each level to the overall evaluation of the productSlide25
Decision Models -- Prof. Juran
25
Product Design ExampleSlide26
Decision Models -- Prof. Juran
26
Product Design Example
Assume that a consumer's purchase decision on an electric razor is based on four attributes, each of which can be set at one of three levels (1, 2, or 3).
Using conjoint analysis, our analysts have divided the market into five segments (labeled as customers 1, 2, 3, 4, and 5) and have determined the "part-worth" that each customer gives to each level of each attribute. Slide27
Decision Models -- Prof. Juran
27
We assume here that all customers within a particular segment view electric razors more or less the same in terms of which levels of which attributes constitute an attractive product.
We also assume that customers in a segment conduct a sort of mathematical analysis (perhaps unconsciously) in which they weigh the various attributes of a product to come up with an overall value with respect to competing products.
Conjoint analysis usually assumes the customer buys the product yielding the highest total part-worth. Slide28
Decision Models -- Prof. Juran
28Slide29
Decision Models -- Prof. Juran
29
For example, consider Segment 1 and these two products:
We assume that customers in Segment 1 will not buy Product B, because they value Product A at 1 + 4 + 4 + 4 = 13 and Product B at 1 + 1 + 1 + 2 = 5.Slide30
Decision Models -- Prof. Juran
30Slide31
Decision Models -- Prof. Juran
31
Currently there is a single product in the market that sets all four attributes equal to 1 (call it Razor 0). We want to introduce two new types of electric razors, and capture as much of the market as possible.
We want to design a two-product line that maximizes the number of market segments that will buy one of our two products. Assume that in the case of a tie, the consumer does not purchase our product.Slide32
Decision Models -- Prof. Juran
32
Managerial Formulation
Decision Variables
Which levels of each attribute to design into each of our two products.
Objective
Maximize the number of customer segments who will buy one of our products.
Constraints
There are only four attributes, each of which must be assigned to one of three existing levels for each product. (In other words, no product can have more or less than one level per attribute.)Slide33
Decision Models -- Prof. Juran
33
Formulation: Preliminaries
There are five customer segments, and we will index them from 1 to 5 with the subscript letter
i
.
There are three products (the existing Razor 0, plus our Razors 1 and 2 to be designed), and we will index them from 1 to 3 with the subscript letter
j
.
There are four product attributes, and we will index them from 1 to 4 with the subscript letter
k
.
There are three possible levels for each attribute, and we will index them from 1 to 3 with the subscript letter
l
.Slide34
Decision Models -- Prof. Juran
34
Symbol
Variable Description
ij
x
A binary variable; 1 if segment
i
will buy product
j
,
0 otherwise.
ij
v
The total “value” that segment
i
places on product
j
.
For our two products (
j
= 1, 2),
1
=
ij
x
if
j
i
ij
v
v
not
products
>
jkl
a
A binary variable; 1 if product
j
has level
l
of
attribute
k
, 0 otherwise.
There are 1
2 of these per product; 36 total in this
problem.
ikl
b
The “value” placed by segment
i
on level
l
of
attribute
k
.
There are 12 of these per segment; 60 total for this
problem (as shown in the table on
slide 27
).
Slide35
Decision Models -- Prof. Juran
35
Example:
Using the example on slide 28, consider Segment 1’s evaluation of Razors A and B.
For Razor A:
Slide36
Decision Models -- Prof. Juran
36
For Razor B:
Since , , and .
In English, customer segment 1 will buy Razor A and not buy Razor B.Slide37
Decision Models -- Prof. Juran
37
FormulationSlide38
Decision Models -- Prof. Juran
38
Solution MethodologySlide39
Decision Models -- Prof. Juran
39
Solution MethodologySlide40
Decision Models -- Prof. Juran
40Slide41
Decision Models -- Prof. Juran
41
Optimal SolutionSlide42
Decision Models -- Prof. Juran
42
Conclusions
It turns out that there is a line of two products that can capture all five segments!
Razor 1, with attribute levels (1, 1, 3, 1), captures segments 4 and 5.
Razor 2, with attribute levels (3, 1, 1, 1), captures segments 1, 2, and 3.Slide43
Decision Models -- Prof. Juran
43
Summary
Evolutionary Solver
(Genetic Algorithm)
Advertising Example
Integer and Binary tricks
Moving Constraints into the productive Function
MAX, IF, SUMIF
Product Design Example
Conjoint Analysis
VLOOKUP, MAX, IF