Upload
norah
View
35
Download
0
Tags:
Embed Size (px)
DESCRIPTION
A BBN as it Would be Developed and Used in the QUELCE Method. Software Engineering Institute Carnegie Mellon University Pittsburgh, PA 15213 Robert W. Stoddard Jim McCurley 24 October 2013. Introduction. - PowerPoint PPT Presentation
Citation preview
2013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
A BBN as it Would be Developed and Used in the QUELCE MethodSoftware Engineering InstituteCarnegie Mellon UniversityPittsburgh, PA 15213
Robert W. StoddardJim McCurley
24 October 2013
32013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Introduction
QUELCE (Quantifying Uncertainty for Early Lifecycle Cost Estimation) is a multi-year research project led by the Software Engineering Measurement and Analysis (SEMA) team within the SEI Software Solutions Division.
Research team membership comprises SEI technical staff with cost estimation background in collaboration with several external faculty (Dr. Ricardo Valerdi, Univ of Arizona, & Dr. Eduardo Miranda, CMU).
This research is motivated by (1) the WSARA Act requiring cost estimates pre-Milestone A and (2) DoD’s need for more accurate cost estimation methods that provide continuous monitoring of changing assumptions and constraints.
42013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Modeling Uncertainty
Complexity Reduction
1. Use QUELCE
Repository to Populate Driver State
Matrix
3. Develop BBN Model and
Assign Conditional
Probabilities to BBN Model
4. Calculate Cost Factor
Distributions for Program Execution
Scenarios
5. Monte Carlo Simulation to Compute Cost
Distribution
2. Evaluate Cause and Effect
Relationships and Reduce Explosion via Dependency Structure Matrix
Overview of QUELCE
Legend:
QUELCE Change Repository
Queries of Historical MDAP Experience and
Context
Change Driver Nominal State Alternative States
Scope Definition
Stable Users added Additional (foreign) customer
Additional deliverable (e.g. training & manuals)
Production downsized
Scope Reduction (funding reduction)
Mission / CONOPS defined New condition New mission New echelon Program
becomes Joint
Capability Definition
Stable Addition Subtraction Variance Trade-offs [performance vs affordaility, etc.]
Funding Schedule
Established Funding delays tie up resources {e.g. operational test}
FFRDC ceiling issue
Funding change for end of year
Funding spread out
Obligated vs. allocated funds shifted
Advocacy Change
Stable Joint service program loses particpant
Senator did not get re-elected
Change in senior pentagon staff
Advocate requires change in mission scope
Service owner different than CONOPS users
Closing Technical Gaps (CBA)
Selected Trade studies are sufficient
Technology does not achieve satisfactory performance
Technology is too expensive
Selected solution cannot achieve desired outcome
Technology not performing as expected
New technology not testing well
● ~~~~ ~~~~ ~~~~ ● ~~~~ ~~~~ ~~~~ ~~~~ ● ~~~~ ~~~~ ~~~~ ~~~~ ~~~~ ~~~~
1. Driver State Matrix
Change Drivers - Cause & Effects Matrix
Mis
sion
/ C
ON
OPS
Cha
nge
in S
trat
egic
Vis
ion
Cap
abilit
y D
efin
ition
Adv
ocac
y C
hang
e
Clo
sing
Tec
hnic
al G
aps
(CBA
)
Build
ing
Tech
nica
l Cap
abilit
y &
Cap
acity
(CB
A)
Inte
rope
rabi
lity
Sys
tem
s D
esig
n
Inte
rdep
ende
ncy
Func
tiona
l Mea
sure
s
Sco
pe D
efin
ition
Func
tiona
l Sol
utio
n C
riter
ia (m
easu
re)
Fund
ing
Sche
dule
Acq
uisi
tion
Man
agem
ent
Prog
ram
Mgt
- C
ontr
acto
r Rel
atio
ns
Proj
ect S
ocia
l / D
ev E
nv
Pro
g M
gt S
truc
ture
Man
ning
at p
rogr
am o
ffice
Mission / CONOPS 3 3 0 6 0Change in Strategic Vision 3 3 3 2 2 2 2 2 3 2 3 2 29 0Capability Definition 3 0 2 1 1 0 0 2 2 2 0 1 0 2 0 0 16 0Advocacy Change 2 1 1 1 1 6 0Closing Technical Gaps (CBA) 2 1 3 1 2 2 1 2 2 1 1 2 1 0 2 2 1 1 2 2 1 2 34 0Building Technical Capability & Capacity (CBA) 1 1 2 1 2 2 1 2 3 2 2 1 2 2 1 1 1 27 0Interoperability 1 2 1 1 1 1 1 1 1 1 2 1 1 2 1 1 3 1 2 2 2 29 1Systems Design 1 2 2 2 2 1 1 1 1 1 2 2 3 21 3Interdependency 1 2 2 1 1 1 1 1 1 1 1 1 2 1 2 2 1 1 1 1 1 2 2 3 33 5Functional Measures 2 2 2 1 1 1 1 1 1 1 2 1 16 0Scope Definition 1 1 3 5 0Functional Solution Criteria (measure) 1 2 2 1 1 2 1 10 1Funding Schedule 1 1 2 1 5 0Acquisition Management 1 1 2 3 1 1 2 2 1 1 1 2 1 19 2Program Mgt - Contractor Relations 1 1 2 1 1 1 1 2 2 12 2Project Social / Dev Env 1 1 1 2 2 1 1 2 1 1 1 14 2Prog Mgt Structure 1 2 1 2 6 1Manning at program office 2 1 2 5 2Scope Responsibility 1 1 1 1 1 1 6 5Standards/Certifications 1 1 1 1 1 1 3 1 10 2Supply Chain Vulnerabilities 1 1 1 1 2 1 7 4Information sharing 1 1 1 1 1 1 1 7 3PO Process Performance 2 2 4 0Sustainment Issues 0 0Contract Award 0 0Production Quantity 2 2 0Data Ownership 2 2 0Industry Company Assessment 0 0Cost Estimate 0 0Test & Evaluation 0 0Contractor Performance 2 2 0Size 0 0Project Challenge 0 0Product Challenge 0 0Totals 0 0 6 4 1 9 5 12 8 7 7 13 4 10 15 18 7 7 8 8 14 17 17 15 12 9 10 13 11 20 19 5 5 17 0Below diagonal 0 0 0 1 1 4 4 4 1 2 0 3 1 3 2 2 3 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Effects
Causes
2. Dependency Structure Matrix
3. BBN Model
4. Cost Factor Distributions by Scenario of Change
5. Monte Carlo with Cost
Estimation Tools
Drivers XL VL L N H VH XH Product ProjectScale Factors
PREC 6.20 4.96 3.72 2.48 1.24 0.00 <X>FLEX 5.07 4.05 3.04 2.03 1.01 0.00 <X>RESL 7.07 5.65 4.24 2.83 1.41 0.00 <X>TEAM 5.48 4.38 3.29 2.19 1.10 0.00 <X>PMAT 7.80 6.24 4.68 3.12 1.56 0.00 <X>
Effort MultipliersRCPX 0.49 0.60 0.83 1.00 1.33 1.91 2.72 XRUSE 0.95 1.00 1.07 1.15 1.24 XPDIF 0.87 1.00 1.29 1.81 2.61 XPERS 2.12 1.62 1.26 1.00 0.83 0.63 0.50 <X>PREX 1.59 1.33 1.12 1.00 0.87 0.74 0.62 <X>FCIL 1.43 1.30 1.10 1.00 0.87 0.73 0.62 <X>SCED 1.43 1.14 1.00 1.00 1.00 <X>
SRDR SARs
52013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
What is a Bayesian Belief Network (BBN)?
A probabilistic model shown in a graphical form consisting of nodes and arrows
Nodes represent factors
Arrows between factors represent cause-effect relationships (ideally), or correlated relationships (minimally)
Factors may be set at “observed” levels representing observed “evidence”
BBNs use traditional conditional probability and Bayesian calculations to update all “unknown” factors based on the latest “evidence”
62013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Burglar Alarm Example of a BBN - 1
Reproduced from Jensen 1996, as seen at http://www.agenarisk.com/resources/example_models.shtml
The following example of a BBN was derived from an example model shown on the AgenaRisk tool vendor website.
This example helps to concisely articulate the operation and use of a BBN.
72013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Burglar Alarm Example of a BBN - 2
(Re-produced from Jensen 1996, as seen at http://www.agenarisk.com/resources/example_models.shtml)
The probabilities of this baseline model reflect both historical data and expert belief.
82013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Burglar Alarm Example of a BBN - 3
(Re-produced from Jensen 1996, as seen at http://www.agenarisk.com/resources/example_models.shtml)
“Mr. Holmes is working at his office when he receives a telephone call from Watson who tells him that Holmes’ burglar alarm has gone off.”
Convinced that a burglar has broken into his house (alarm sounds -> burglary), Holmes rushes into his car and heads for home.”
92013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Burglar Alarm Example of a BBN - 4
(Re-produced from Jensen 1996, as seen at http://www.agenarisk.com/resources/example_models.shtml)
“On his way he listens to the radio, and in the news it is reported that there has been a small earthquake in the area (radio report -> earthquake). Knowing that the
earthquake has a tendency to turn the burglar alarm on (earthquake -> alarm sounds), he returns to his work leaving his neighbors the pleasures of the noise.”
102013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Example QUELCE Bayesian Belief Network
112013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Benefits of Using a BBN Model for QUELCE – 1
Models the uncertainty of program change drivers and their relationships using probability distributions.•No longer use single point estimates• Instead, we use ranges and distributions reflecting uncertainty
Provides continuous measurement and ability to update and re-estimate based on changes in program execution.•Once created, simple to run scenarios as new “evidence” is observed•Readily used to update cost estimates based on changing program conditions
Translates the net effect of program change driver uncertainty to the input factors of cost estimation models.•We use Monte Carlo simulation to translate BBN output distributions into
distributions for input parameters of CERs
122013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Benefits of Using a BBN Model for QUELCE – 2
Enables the use of both objective (hard) information and subjective (soft) information as evidence to update our forecast.•Evidence can be factual observation, e.g. something has happened •Evidence can be subjective in terms of a person’s anticipation of an event
occurring
Enhances ability to conduct “what-if” analysis in context of change drivers.•A scenario in the BBN is a collection of one or more change drivers observed
or anticipated to occur with a specified probability•For each scenario, the BBN then recalculates and produces new outcome
node distributions
132013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Benefits of Using a BBN Model for QUELCE – 3
Enables analysis and forecast with incomplete information (e.g., no status available for some change drivers).•Traditional statistical analysis requires entries for all modeled factors, e.g. in a
regression equation•BBNs can provide updated assessments of all unknown factors based on
whatever factors are observed
Provides the ability to determine which change drivers are most influential on downstream change drivers or BBN outcome nodes (e.g., project complexity, product complexity).•Any factor may be selected for evaluating sensitivity to any and all other factors•For any factor, a sensitivity “tornado” chart may be created depicting in
descending order all other factors influencing this factor
142013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Benefits of Using a BBN Model for QUELCE – 4
May explain the most likely state of affairs of upstream change drivers based on current observations of downstream change drivers.•Akin to diagnosing likely causes of today’s observations•Commonly used in medical diagnosis•May be used to diagnose what other change drivers led to the current state of
affairs of known change driver occurrence
152013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Future Work
Need to standardize the output nodes of the QUELCE BBN
Need to provide a method to connect the BBN output nodes to the inputs of commonly used CERs
Need to collect data from retrospectives and ongoing program executions to validate the performance of the QUELCE BBN and method
162013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Summary
The DSM matrix captures the experts’ change probabilities for one change driver affecting another change driver (all possible pairings).
The BBN models the probabilistic relationships so that different change scenarios including cascading change may be evaluated.
For each scenario, the BBN produces probability distributions for the output nodes which will then be used to assign probability distributions to the input factors of the cost estimating relationships/tools.
172013 COCOMO ForumStoddard, 24 October 2013© 2013 Carnegie Mellon University
Contact Information
Robert W. StoddardPrincipal ResearcherSoftware Solutions Division, SEAPTelephone: +1 412-268-1121Email: [email protected]
U.S. MailSoftware Engineering InstituteCustomer Relations4500 Fifth AvenuePittsburgh, PA 15213-2612USA
Webwww.sei.cmu.eduwww.sei.cmu.edu/contact.cfm
Customer RelationsEmail: [email protected]: +1 412-268-5800SEI Phone: +1 412-268-5800SEI Fax: +1 412-268-6257