Planning III Estimating Software Size

1
Topic 5

• Planning III Estimating Software Size
• ??? ?????? ???

2
Lecture Overview

• 5.1 Background
• 5.2 Popular Estimating Methods
• 5.3 Proxy-based Estimating
• 5.4 The PROBE Size Estimating Method
• 5.5 Object Categories
• 5.6 Estimating Considerations
• 5.7 Summary
• 5.8 Exercises

3
5.1 Background
4
5.1 Background 1

• Why Estimate Size?
• Size estimates allow you to make better plans.
• Plan better for the resources, tasks, and
  schedule.
• The quality of a plan depends on that of the size
  estimate.
• The more details you have, the more accurate the
  estimation will be.
• For a large software job, divide it into separate
  elements.
• Size estimates assist you in tracking progress.
• You can
• judge when the job scope changes.
• better measure the work.

5
Background 2

• Estimating models in other fields
• have a large historical base.
• are widely used.
• generate detailed planning data.
• require a size estimate as input.

6
Project Planning Framework

• The framework is shown in Figure 5.1.
• Size estimating compare the design elements with
  the historical size data to make estimates.
• Accurate size estimation will lead to accurate
  resource estimation.

7
Fig. 5.1 Project Planning Framework
8
Estimating Experience

• Studies show that
• size estimation errors may be as high as 100 or
  even more.
• Only 22 professionals make estimates.
• Usually people have a larger estimation error
  percentage (up to 400) in early phase, and then
  the percentage declines (25 or less) in later
  phases.
• Serious size estimation errors result in poor
  resource estimates and unrealistic project
  schedules.
• Fortunately, the size estimation skill can be
  learned and improved.

9
Size Estimating Criteria

• A widely used method should be
• structured and trainable.
• usable during all development and maintenance
  phases.
• usable for all types of software product
  elements.
• suitable for statistical analysis.
• adaptable to the types of your future work.
• able to judge the accuracy of the estimates.

10
5.2 Popular Size Estimating Methods
11
5.2 Popular Size Estimating Methods

• Four popular methods
• Wideband-Delphi
• Fuzzy-Logic
• Standard-Component
• Function-Point
• Their concepts form the foundation of the PROBE
  (Proxy-based Estimating) method used with the PSP.

12
Wideband-Delphi Method 1

• Originated by Rand Co. and refined by Boehm.
• Based on the Delphi process, which usually
  contains several cycles.
• Several experts (estimators) and a moderator are
  involved in the process.
• Each expert makes an independent estimate and
  submit it to the moderator in each cycle.
• The moderator coordinates the estimation process
  for these experts whose estimates are anonymous
  to all others.
• The process runs until experts estimates
  converge on a consensus result.

13
Wideband-Delphi Method 2

• The methods process is as follows
• A group of experts is each given the programs
  specifications and estimation forms.
• They meet to discuss project goals, assumptions,
  and estimation issues.
• They then each anonymously list project tasks and
  estimation size.
• The estimates are given to the moderator, who
  tabulates the results and returns them to the
  experts, as illustrated in Figure 5.2.

14
Fig. 5.2 The Chart Used in Wideband-Delphi
15
Wideband-Delphi Method 3

• Only each experts personal estimate is
  identified all others are anonymous.
• The experts meet to discuss the results. They
  each review the tasks they have defined but not
  their size estimates.
• The cycle continues at step 3 until the estimates
  converge to within an acceptable range.

16
Wideband-Delphi Method 4

• Advantages
• can produce accurate results
• uses organizations skills
• works for any size products.
• Disadvantages
• relies on a few experts
• time consuming
• subject to common biases

17
Fuzzy-Logic Method 1

• Estimators compare the planned program with prior
  programs and select the most appropriate size
  category.
• Preparing for this method
• gather sufficient size data on previously
  developed programs
• subdivide these data into size categories and
  subcategories
• provide a meaningful number of program examples
  in each size category and subcategory
• extend the size ranges up or down as you get
  further data
• Table 5.1 shows the example.

18
Table 5.1 Example for Fuzzy-Logic Method
19
Fuzzy-Logic Method 2

• Advantages
• based on relevant historical data
• easy to use
• requires no special tools or training
• provides reasonably good estimates in cases where
  new work is like prior experience

20
Fuzzy-Logic Method 3

• Disadvantages
• requires a lot of data
• requires that the estimators be familiar with the
  historically developed programs
• provides only a crude sizing
• not useful for new program types
• not useful for programs that are much larger or
  smaller than the historical data

21
Standard-Component Method 1

• Use organizations historical data to determine
  the typical size of various types of components
• subsystems, modules, screens, reports, etc.
• For a new project
• Judge the number of each type of components will
  likely be in the project.
• Also determine the maximum and minimum numbers of
  each type of components you could image.

22
Standard-Component Method 2

• Calculate the estimated number of each type with
• (4likely number max. number min. number) / 6
• The estimated size of each type is
• its estimated number its typical size from
  historical data
• Sum up the estimated size of all types to get the
  total size.

23
Table 5.2 Example of Component Estimating
For modules, its estimated size 932 17.5
16310.
24
Standard-Component Method 3

• Advantages
• based on relevant historical data
• easy to use
• requires no special tools or training
• provides a rough estimate range
• Disadvantages
• must use large components early in a project
• limited data on large components
• hard to visualize component counts early in a
  project

25
Function-Point Method 1

• Count the numbers of the 5 types of basic
  functions that the commercial application program
  will likely need by reviewing its requirements.
• Inputs screens or forms through which to add new
  data or update existing data in the application
• Outputs screens or reports that the application
  produces
• Inquires screens used to ask for interrogation,
  assistance or information
• Data files logical collection of records that
  the application updates
• Interface e.g., files shared with other
  applications, shared databases, parameter lists,

26
Function-Point Method 2

• Each type has a given weight.
• Calculate the function points of the application
• multiply the number of each type by its weight.
• sum them up to get the total (unadjusted).
• Use historical data on development cost and time
  per function point to make the estimates.

27
Table 5.3 Function-Point Categories
28
Table 5.4 Function-Point Category Example
29
Function-Point Method 3

• To improve the estimation, adjust the function
  points by 14 influence factors.
• The influence factor values are selected from 0
  (very simple) to 5 (very complex) by the judgment
  of the estimator.
• Sum up these values.
• Calculate the complexity multiplier
• 0.650.01(sum of influence factor values)
• Calculate adjusted function points
• complexity multiplier unadjusted function
  points
• The adjustment is between 35
• Table 5.5 shows an example.

30
Table 5.5 An Example of Function-Point Influence
Factors
31
Function-Point Method 4

• Advantages
• a well-documented method
• usable in the earliest requirements phase
• independent of programming languages, product
  designs, development styles
• having a large body of historical data
• with an active user group

32
Function-Point Method 5

• Disadvantages
• cannot directly count an existing products
  function point content
• without historical data, difficult to improve
  estimating skill
• not reflect language, design, and style
  differences
• designed for estimating commercial data
  processing applications

33
5.3 Proxy-based Estimating (PROBE)
34
5.3 Proxy-Based Estimating (PROBE)

• In stead of direct estimating, a proxy is a
  substitute to help estimators judge product size.
• Proxies, for example, can be objects, screens,
  files, scripts, document chapters, function
  points, and so on.

35
Criteria for a Good Proxy 1

• The proxy size measurement (or estimate) should
  closely relate to the effort required to develop
  the product.
• Use the correlation method (see Topic 15) to
  determine which proxy is a better predictor of
  product size or development effort.

36
Criteria for a Good Proxy 2

• The proxy content of a product should be
  automatically countable.
• Large amount of proxy data extracted from
  historical data is needed to define new
  estimates.
• The proxy must be a physical entity that can be
  precisely defined and algorithmically identified.

37
Criteria for a Good Proxy 3

• The proxy should be easy to visualize at the
  beginning of a project.
• The proxy should be customizable to the special
  need of using organizations.
• Different product types may use different kind of
  proxies to estimate.
• The proxy should be sensitive to any
  implementation variations that impact development
  costs or efforts.
• for example, program languages, design styles,
  application types

38
Objects as Proxies 1

• Objects in OO programming languages are a good
  candidate as proxies because they closely relate
  to development effort or resources.
• Figures 5.3 and 5.4 show close correlation
  between estimated object LOC (Sobject size ) and
  actual program LOC.
• Linear regression formula (See Topic 15 )
• actual program LOC ß0 estimate object
  LOC ß1
• Figures 5.5 and 5.6 show high correlation and
  significance between estimated object LOC and
  actual development hours.
• Linear regression formula
• actual development hours ß0 estimate
  object LOC ß1

39
Objects as Proxies 2

• The PROBE method uses objects as proxies.
• The PROBE method requires that estimators have
  historical data on the sizes of objects they have
  developed and that these data be divided into
  categories.

40
Fig. 5.3 Estimated Object LOC vs. Actual Program
LOC (10 Pascal Programs)
41
Fig. 5.4 Estimated Object LOC vs. Actual Program
LOC (25 C Programs)
42
Fig. 5.5 Estimated Pascal Object LOC vs. Actual
Development Hours with Correlation
0.934, Significance lt 0.005
43
Fig. 5.6 Estimated C Object LOC vs. Actual
Development Hours with Correlation
0.980, Significance lt 0.005
44
Estimating Object Size with Fuzzy-Logic Method

• Judge the size of each object on a per-objects
  method basis with fuzzy-logic method.
• decide the category of the object
• judge how many methods it likely will contain
• decide the size range it falls into
• estimated object size ( of methods) (LOC of
  the size range)
• Table 5.6 shows object category sizes in LOC
  per-objects method
• For example, a medium-sized Pascal text object
  with 4 methods has about 66 (16.484) LOC.

45
Table 5.6 Object Category Sizes in LOC per Method
46
5.4 The PROBE Size Estimating Method
47
5.4 The PROBE Size Estimating Method

• The PROBE size estimating procedure is shown in
  Figure 5.7.
• In the procedure, a size estimating template,
  shown in Table 5.7, will be used.
• Use the template to instruct how to estimate
  program size with the PROBE method and linear
  regression.

48
Fig. 5.7 The Procedure of PROBE Method
49
Table 5.7 Size Estimating Template and example 1
50
Table 5.7 Size Estimating Template and example 2
51
Size Estimating Template 1

• Base Program the program you will enhance and
  perform any changes to it.
• Base Size (B) the LOC of the base program
• LOC Deleted (D) the LOC to be deleted from the
  base program
• LOC Modified (M) the LOC to be modified in the
  base program

52
Size Estimating Template 2

• Projected LOC (P) includes total base additions
  LOC (BA) and total new objects LOC (NO)
• Base Additions the new functions to be added to
  the base program
• New Objects the new objects to be added to the
  base program
• New Reused Objects a new object planned to
  develop and general enough to put into the reuse
  library
• (Note mark an with the new reused objects)
• Reused Objects (R) the objects taken from the
  reuse library

53
Size Estimating Template 3

• Calculations
• Using actual new and changed LOC and estimated
  Object LOC (i.e., PM) from historical data to
  obtain regression parameters ß0, ß1
• Estimated New and Changed LOC (N) use regression
  parameters to calculate
• N ß0 ß1 (P M)
• Estimated Total LOC (T) the estimated size of
  the final program

54
Size Estimating Template 4

• Prediction Range using t distribution (See
  Section 15.5.2), a desired prediction interval
  percent, and the data for linear regression to
  calculate
• Lower/Upper Prediction Interval the interval (N
  Range) within which the actual new and changed
  LOC is likely to fall
• Prediction Interval Percent the percentage that
  the actual new and changed LOC is likely to fall
  within the interval

55
PROBE Step 1 Conceptual Design

• The conceptual design establishes a preliminary
  design approach and names the expected product
  objects and their functions.
• For an accurate estimate, estimators must refine
  the conceptual design to the level of objects.

56
PROBE Step 2 Identify Objects

• Determine object type and size
• if it is a new object, use fuzzy-logic method
  (see Section 5.3)
• of methods (judging size using a per-method
  basis)
• object type
• relative size
• LOC per method
• estimated object size ( of method) (LOC per
  method)
• (Note mark an with the new reused objects)
• if the object is taken from the reuse library,
  determine
• reuse object categories
• object LOC

57
PROBE Step 3 Calculate Projected and Modified
LOC

• Projected LOC (P)
• Total Base Additions LOC (BA)
• Total New Objects LOC (NO)
• Using actual new and changed LOC and estimated
  object LOC (i.e., PM) from historical data to
  obtain regression parameters ß0, ß1
• Estimated New and Changed LOC (N)
• ß0 ß1 (PM)

58
PROBE Step 4 Estimate Program Size

• Estimated program size
• Base Size (B)
• Estimated New and Changed LOC (N)
• Reused Total LOC (R)
• LOC Deleted (D)
• LOC Modified (M)
• LOC Modified (M) is subtracted because it is
  counted twice one in Base Size (B) and the other
  in Estimated New and Changed LOC (N).

59
PROBE Step 5 Calculate Prediction Interval 1

• The purpose is to assess the quality of the
  estimate.
• Use t distribution, a desired prediction interval
  percent, and the data for linear regression
  calculation to get
• Prediction Range
• Prediction Interval LPI, UPI
• Within the prediction interval, the actual new
  and changed LOC is likely to fall with the
  selected percent.

60
PROBE Step 5 Calculate Prediction Interval 2

• Prediction Interval LPI, UPI
• Lower Prediction Interval (LPI)
• Estimated New and Changed LOC (N) -
  Range
• Upper Prediction Interval (UPI)
• Estimated New and Changed LOC (N)
  Range

61
PROBE Step 5 Calculate Prediction Interval 3

• Prediction Range using the data for linear
  regression calculation in Step 3
• Where
• xi is the estimated object LOC (i.e., PM) in
  historical data
• xavg is the average of xi
• xk is the estimated object LOC (i.e., PM) in the
  new program
• yi is actual new and changed LOC in historical
  data

62
5.5 Object Categories
63
5.5 Object Categories

• The PROBE method uses objects as proxies. It
  needs object categories and their sizes to judge
  the size of the new objects.
• Table 5.6 shows the example that we want to
  produce.
• for C and Object Pascal
• object sizes per method are divided into 5 size
  ranges very small, small, medium, large, and
  very large.

64
Object Size Distribution

• Assume historical object data are normally
  distributed.
• Figure 5.8 shows the relation between the normal
  distribution and the size ranges.

65
Fig. 5.8 Normal Distribution with Size Ranges
s
66
Approaches

• Two approaches
• If the size data are close to normal
  distribution, use normal distribution in the
  calculation .
• Otherwise, use normal distribution of natural log
  in the calculation.

67
Approach I Procedure with Original Data

• Use historical data on objects
• divide them into categories
• calculate the size range midpoints for each
  category.
• use statistical techniques of normal distribution
  in the calculation
• The following is the procedure

68
Step 1 Classification

• Divide objects in historical data into
  categories.
• Basic object categories
• logic, control
• I/O, files, display
• data, text, calculation
• set-up, error handling

69
Step 2 Calculating LOC per Method

• For each category, calculate the LOC per method
  of each object.
• Table 5.9 shows the example for 13 objects
  belonging to the object category, text.

70
Table 5.8 Pascal Text Object LOC per Method
71
Step 3 Calculating Standard
Deviation

• For each category, calculate the variance and
  standard deviation of the values from Step 2.
• Variance s2 (1/n) ?i1..n(xi xavg )2
• Standard Deviation v of Variance s
• where xi (LOC per method) of each object of
  one specific category xavg denotes the average.
• The following table shows the example.

72
Table 5.9 Pascal Text Object Standard Deviation
73
Step 4 Calculating Size Range
Midpoints

• For each category, use the average xavg and
  standard deviation s to calculate the size range
  midpoints
• VL xavg 2s
• L xavg s
• M xavg
• S xavg s
• VS xavg - 2s

74
Negative Size Range Midpoints

• Table 5.10 shows the example of the size range
  midpoints and the 13 object sizes per method.
• However, in this example, the VS midpoint is a
  negative number, which is not what we want and
  means that the object data are not normally
  distributed.

75
Table 5.10 Pascal Text Object and Size Ranges
76
Approach II Procedure with Natural Log of
Original Data

• The trick to handle the negative size range
  midpoint is to calculate the natural log of the
  data.
• The following is the procedure with the first 2
  steps same as the previous procedure.

77
Step 3 Calculating the Natural Log
and Average

• For each category, calculate the natural log (ln)
  of the LOC per method of each object, and the
  average, avgln, of these log values.
• Table 5.11 is the example for object category,
  text.

78
Table 5.11 Pascal Text Object ln (LOC per Method)
79
Step 4 Calculating Standard
Deviation

• For each category, calculate the variance and
  standard deviation of these log values.
• Variance s2 (1/n) ?i1..n(xi xavg )2
• Standard Deviation v of Variance s
• where xi ln (LOC per method) of each object
  of a category.

80
Step 5 Calculating the Log of Size
Range Midpoints

• For each category, use average of log values and
  standard deviation to calculate the log of size
  range midpoints
• ln(VL) avgln 2s
• ln(L) avgln s
• ln(M) avgln
• ln(S) avgln s
• ln(VS) avgln - 2s

81
Step 6 Calculating the Antilog

• For each category, calculate the antilog to get
  the midpoints of the size ranges
• VL eln(VL)
• V eln(L)
• M eln(M)
• S eln(S)
• VS eln(VS)
• Table 5.12 shows the example of the size range
  midpoints and the 13 object sizes per method.

82
Table 5.12 Pascal Text Objects and Size Ranges
83
5.6 Estimating Considerations
84
5.6 Estimating Considerations 1

• The PSP estimating objective is to improve your
  estimating ability by tracking and analyzing your
  estimates.
• If your estimating process is stable, the linear
  regression method, based on the historical data
  gathered from consistently biased estimates, can
  make an accurate bias adjustment.
• While you gain more experience and your process
  evolves, you should adjust your statistical
  calculations to include only the newer and more
  representative data and drop the old ones.

85
Estimating Considerations 2

• Whenever your linear regression parameters appear
  unreasonable (e.g., ß1 far from 1.0, large ß0 ),
  use the averaging method to estimate.
• This method uses a ratio to adjust size or time
  based on historical averages.
• estimated program LOC estimated object LOC
  ratio
• where ratio historical average
• ?(final program LOC) / ? (estimated
  object LOC)

86
Estimating Considerations 3

• To use the linear regression method, you must
  have at least three programs for which you have
  made object LOC estimates.
• If you have at least three programs but enough
  historical estimate data,
• obtain ß0 and ß1 from your actual size data
• If you do not have enough historical data but at
  least one program,
• use the averaging method

87
Estimating Considerations 4

• You can learn to reduce the estimating bias by
  comparing, during postmortem, estimates made at
  each phase with their actual size.
• To estimate unprecedented products, the best
  answer is to resist making firm estimates until
  you have completed a feasibility study and build
  some prototypes.

88
5.7 Summary
89
5.7 Summary 1

• Accurate size estimate will help you to make
  better development plans.
• Size estimating skill improve with practice.
• A defined and measured process provides a
  repeatable basis for improvement.

90
Summary 2

• With PROBE, estimates are based on one of
  following four methods
• Method A regression with estimated object LOC
• Method B regression with estimated new and
  changed LOC
• Method C size or time adjustment based on
  historical averages (the averaging method)
• Method D engineering judgment

91
Summary 3

• Method A regression with estimated object LOC
• use the relationship between estimated object LOC
  and
• actual new and changed LOC
• actual development time
• The criteria for using this method are
• 3 or more data points that are correlated (r2 gt
  0.5)
• Reasonable regression parameters

92
Summary 4

• Method B regression with estimated new and
  changed LOC
• use the relationship between estimated new and
  changed LOC and
• actual new and changed LOC
• actual development time
• The criteria for using this method are
• 3 or more data points that are correlated (r2 gt
  0.5)
• Reasonable regression parameters

93
Summary 5

• Method C adjusting size or time based on
  historical averages.
• the averaging method
• easy to use
• used when linear regression parameters (ß0, ß1)
  appear unreasonable (that is, methods A and B can
  not be used)
• assuming that there is no fixed overhead
• The criteria for using this method are
• At least one data point is required.

94
Summary 6

• Method D engineering Judgment with estimated
  object LOC
• used in absence of historical data
• using judgment from estimated object LOC to
  estimate
• actual new and changed LOC
• actual development time
• used when methods A, B, and C can not be used

95
5.8 Exercises
96
Program 3A 1

• Use PSP0.1 (see Appendix in Topic 4) to write
  program 3A.
• Program 3A Requirements
• Write a program to count
• the total program LOC
• the total LOC in each object
• the number of methods in each object
• If an object-oriented program is not used as the
  input, count
• the total program LOC
• the total LOC in each procedure or function

97
Program 3A 2

• Use the counting standard produced by report
  exercise R1.
• It is acceptable to enhance program 2A or to
  reuse some of its methods, procedures, or
  functions in developing this program.
• Program 3A Testing
• Thoroughly test the program.
• At a minimum, test program 1A, 2A, and 3A.
• Prepare a test report that includes a table as
  the format in Table 5.11.

98
Table 5.11 Test Result Format ? Program 3A
a) Format for object-oriented program designs
b) Format for non object-oriented program designs
99
Program 4A 1

• Use PSP1 (see Appendix in this topic) to write
  program 4A.
• Program 4A Requirements
• Write a program to calculate the linear
  regression size-estimating parameters

100
Program 4A 2

• Use the historical data of a set of n programs
  object LOC, xi, and new and changed LOC, yi.
• Enhance the linked list of program 1A to hold the
  n data records, where each record has 2 real
  numbers.

101
Program 4A 3

• Program 4A Testing
• Thoroughly test the program.
• At a minimum, test it with the 3 cases shown in
  Table 5.12
• use the data for estimated object LOC and actual
  new and changed LOC.
• use the data for estimated new and changed LOC
  and actual new and changed LOC.
• use the data, estimated new and changed LOC and
  actual new and changed LOC, you have gathered for
  the programs 2A, 3A and 4A you have developed.
• Prepare a test report that includes a table as
  the format in Table 5.13.

102
Table 5.12 Size Estimating Regression Data
 
103
Table 5.13 Test Result Format ? Program 4A
104
Appendix PSP 1
105
PSP1 Process Script 1
106
PSP1 Process Script 2
107
PSP1 Planning Script 1
108
PSP1 Planning Script 2
109
PSP1 Development Script 1
110
PSP1 Development Script 2
111
PSP1 Postmortem Script 1
112
PSP1 Postmortem Script 2
113
Size Estimating Template 1
114
Size Estimating Template 2
115
Test Report Template
116
PSP1 Project Plan Summary 1
117
PSP1 Project Plan Summary 2