Title: Ten Keys to Success in Optimization Modeling
1Ten Keys to Success in Optimization Modeling
- Richard E. Rosenthal
- Operations Research Department
- Naval Postgraduate School
INFORMS Atlanta, October 2003
2Theme
- Optimization is valuable and pervasive (no need
to preach to the converted here) - Practical optimization applications continue to
be - bigger
- more complex
- closer to real-time (if the situation warrants)
- less dependent on OR gurus
- more depended upon by companies, and
- taken for granted or
- taken over and claimed credit for by non-ORs
3Theme (continued)
- As we all know, this has been made possible by
remarkable improvements in computers, modeling
systems, and solvers (algorithms and their
implementations). We have many great researchers
and commercial implementers to thank. - But, there is also another, important, sometimes
overlooked piece of the story - GOOD MODELING PRACTICE
4Acknowledgements
- My ideas on good modeling practice have been
greatly influenced by my NPS colleagues Jerry
Brown, Matt Carlyle, Rob Dell, Kevin Wood, and
other great modelers I have observed in the
practice of their art, such as Harlan Crowder,
Terry Harrison, Karla Hoffman, David Ryan, Linus
Schrage, Julie Ward, Andres Weintraub, Kirk Yost. - Many great ideas have come from NPS students.
- Todays talk owes a special debt to Jerry Brown.
5Can You Teach Modeling?
6Key 1 Communicate Early and Often
- Mathematical formulation kept up to date
- Verbal description of formulation
- Executive summary in the right language
7Mathematical Formulation
- Index use
- Given data (and units)
- in lower case
- Decision Variables (and units)
- in UPPER CASE
- Objectives and constraints
8Verbal Description of Formulation
- Constraints 3 ensure that one service facility
is assigned responsibility for each product line
p. - You wonder why I mention this, but look at our
applied literature.
9Non-mathematical Executive Summary
- Jerry Browns Five Essential Steps
- What is the problem?
- Why is the problem important?
- How will the problem be solved without you?
- How will you solve the problem?
- How will the problem be solved with your results,
but without you?
10Refining Your Executive Summary
- Have a non-OR read your summary out loud
- Ask the reader to explain what is going on
- Listen well
- Revise and repeat
- If you dont learn how to speak in the
executives language, then someone less qualified
than you will be entrusted with solving their
problems.
11Non-mathematical Executive Summary
- Thats it? Thats peer review?
12Key 2 Bound all Decisions
- A trivial concept, too-often ignored
- Remember all the formal neighborhood
assumptions underlying your optimization method?
- Bob Bixby tells of real customer MIP with only 51
variables and 40 constraints that could not be
solved until bounds were added, and then it
solved in a flash.
13Bound all Decisions
- Optimization is an excellent way to find data
errors, but it really exploits them - Moderation is a virtue
- Bonus! You never have to deal with the
embarrassment (or the theory) of - unbounded models.
14 Bound all Decisions
15Key 3 Expect any Constraint to Become an
Objective, and Vice Versa
- Real-world models are notorious for multiple,
conflicting objectives - Expert guidance from senior leaders is often
interpreted as constraints - These constraints are often infeasible
- Discovering what can be done changes your concept
of what should be done - Contrary to impression of textbooks, alternate
optima are the rule, not the exception.
16Key 4 Sensitivity Analysis in the Real World Is
Nothing Like Textbook SA
- LP Sensitivity Analysis, Textbook Style
- Disappointing in practice because theory creates
limits. -
- Textbooks have sleek algorithms for one
modification at a time, all else held constant,
e.g., - minimize Sj cj Xj dXk
- Not very exciting in practice. So why is this
stuff in all the textbooks? What is worth
talking about?
17LP Sensitivity, Practitioner Style
Operations Research, Jul-Aug 2002
18Sensitivity Analysis, Practitioner Style
Large-scale LP for optimizing airlift -- multiple
time-space muticommodity networks, linked
together with non-network constraints .
Initial results on realistic scenario only 65
of required cargo can be delivered on time.
Analysis of result revealed most of the
undelivered cargo was destined for City A from
City B, so.. what if we redirect some of this
cargo to City A? Sensitivity analysis add
12,000 new rows and 10,000 columns... on-time
delivery improves to 85.
19Forget Textbook Sensitivity Analysis, Plan on
Lots of Model Excursions
- The beauty of this is that it is only of
theoretical importance, and there is no way it
can be of any practical use whatsoever!
20Key 5 Bound the Dual Variables
- Huh?
- Elastic constraints,
- with a linear (or piece-wise linear) penalty per
unit of violation, - bound the dual variables
- Im willing to satisfy this restriction
(constraint), - as long as it doesnt get too expensive.
- Otherwise, forget it
- Ill deal with the consequences
21Key 6 Model Robustly
- Your analysis should consider alternate future
scenarios, and render a single robust solution. - There may be many contingency plans,
- but you only get one chance per year
- to ask for the money to get ready.
22Model Robustly
- This is the part I always hate.
23Key 7 Eliminate Lots of Variables
- Big models get to be big through Cartesian
products of indices - Find rules for eliminating lots of index tuples
before they are generated in the model - Sources of rules mathematical reasoning and
common sense based on understanding of the
problem - You can often eliminate constraints too!
24Example 1 of Variable Elimination
- XD(a,i,r,t) of type a aircraft direct
delivering cargo for customer i on route r
departing at time t - Allow variable to exist only if
- Route r is a direct delivery route from
customer i s origin to i s destination - Aircraft type a is available at i s origin
at t - Aircraft type a can fly route r s critical
leg - Aircraft type a can carry some cargo type that
customer i demands - Time t is not before i s available-to-load
time - Time t is not after i s required delivery
date maxlate travel time
25Example 2 of Variable Elimination
- Ann Bixby and Brian Downs of Aspen Technology
developed real-time Capable-to-Promise model for
large meatpacking company - One of their major efforts to bring solution
times down low enough was variable elimination.
26Eliminate Lots of Variables
-
- You wont win the Nobel or Lanchester Prize for
this key idea, but it really, really helps.
27Eliminate Lots of Variables
- In effect, what youre doing is taking a big
lead off third.
28Key 8 Incremental Implementation
- In a complex model, add features incrementally.
Test each new feature on small instances and
take no prisoners. - When new features dont work, there is either a
bug to be fixed or a new insight to be gained.
Either way, treasure the learning experiences.
29Incremental Implementation
- Eliminate variables corresponding to airlifters
switching from long-haul to shuttle status, if
there are no foreseeable shuttle opportunities.
Feature tested with small example removing the
option to make a seemingly foolish decision
actually caused degradation of objective
function. What happened?
30Key 9 Persistence
- Any prescriptive model that suggests a plan and
then, when used again, ignores its own prior
advice - is bound to advise something needlessly
different, and lose the faith of its
beneficiaries. - Jerry Brown
31Illustration of Persistence
- There are initially 8 customers to serve. We
must choose serving site and equipment type.
32Illustration of Persistence
- Just a moment after this solution is announced,
two high-priority customers call in. The model
is rerun with the 10 customers. - There are not enough assets to cover all 10
customers. - The new solution requires a major reallocation
of assets. Major changes in the solution are
highlighted.
33Illustration of Persistence
34Illustration of Persistence
- A persistent version of the model is run to
obtain a new optimal solution that discourages
major changes from the original announced
solution. - Add to objective function penalties on
deviations from original solution, weighted by
severity of disruption.
35Illustration of Persistence
36Illustration of Persistence
- At a cost of 1 in the objective function, the
persistent solution causes no disruption to the
announced plans, other than substitution of the
two new customers.
37Persistence
- Key reference
- G. Brown, R. Dell, K. Wood, Optimization and
Persistence, Interfaces 1997. -
38Key 10 Common Sense
- Heuristics are easy --- so easy we are tempted to
use them in lieu of more formal methods -
- Heuristics may offer a first choice to assess a
common sense solution - But, heuristics should not be your only choice
39Common Sense
- A formal optimization model takes longer to
develop, and solve - But it provides a qualitative bound on each
heuristic solution - Without this bound, our heuristic advice is of
completely unknown quality - This quality guarantee is key
40Common Sense
- Its OK to use a heuristic,
- but you should pair it with a traditional,
calibrating mathematical model - With no quality assessment,
- you are betting your reputation
- that nobody else is luckier than you are
41Common Sense
- I tend to agree with you,
- especially since thats my lucky number.
4210 Keys to Success in Optimization Modeling
- 1 Write formulation, communicate with execs
- 2 Bound decisions
- 3 Objectives and constraints exchange roles
(alt. optima likely) - 4 Forget about sensitivity analysis as you
learned it - 5 Elasticize (bound duals)
- 6 Model robustly
- 7 Eliminate variables avoid generating them
when you can - 8 Incremental implementation
- 9 Model persistence
- 10 Bound heuristics with optimization
4310 Keys to Success in Optimization Modeling
- Thats it? Thats the Grand Unified Theory?
44Questions and Comments?
Stephen Hansen, Man on a Limb