Lecture 3
Measurement scales for software metrics
1 Nominal Data: one can measure the type of program by placing it in a category of some kind. E.g. DB, operating system. Here it is impossible to use arithmetic ranking. Only type can be compared!!! Categories
2 Ordinal Data: Here values like medium, high, low can be used. Rankings
3 Interval: Differences
4 Ratio: Absolute Zero (scale 0 onwards)
Some Examples
1 E.g. Ratio scale, program having 2000 LOC (lines of code) can be interpreted as twice as large as program having 1000 LOC. What assumptions are needed?
2 Are these important for the metric to be successful?
Product Metrics
Most work deals with the source code and its characteristics. Also the complexity of the code is something that can be examined!
LOC
1 Most widely used and simplest metric 2 Looks simple 3 Problems start when we try to consider blank lines, non-executable
statements, compound statements in a single line!, reused code in program
code etc!!!
4 LOC ignores all the above mentioned features!
5 Halstead’s program length N can offer a better measure!
6 Why do people keep using it???
Product Metrics
1 Size metrics 2 LOC metrics 3 Function points 4 Complexity measures
Function points
1 Created by Albrecht 2 Compute the function points (FP) 3 Based upon the number of features 4 Suitable for OO programming
Original FP
No Function Point Weight
1 External user inputs 4
2 Inquiries 4
3 Outputs 5
4 Master file 10
Each FP can be adjusted within a -/+ 35% for a specific project complexity.
Recent work with FP
1 Many different variants
2 Consider even the complexity of the algorithms being used.
3 More complex than the original FP developed by Albrecht
4. Can be derived from formal specifications.
COCOMO
1 Best known and most documented model
2 Provides three levels of models: basic, intermediate and detailed.
3 Boehm identifies three modes of product development. Organic,
Semidetached, embedded. This enables one to understand the level of difficulty of the project.
4 It is suggested that the detailed model will provide cost estimates that are within 20% of the actual values 70% of the time. Pred(0.20) = 0.70
Welcome back to Kaizenlog.com, you may want to subscribe to my RSS feed , Twitter You can contact us by using the contact form or submitting a comment. You can also share this post with your friends by clicking on the 'ShareThis' button above. Thanks for visiting!
Print This Post
Entries (RSS)