Software Analytics for Pragmatists (DevOps Camp 2017)

Each step in the development or use of software leaves valuable, digital tracks. The analysis of this "software data" (such as runtime measures, log files or commits) refines our gut feeling to facts with sound evidence.

I'll show how questions that arise in software development can be answered automated, data-driven and reproducible. I demonstrate the interaction of open source analysis tools (such as jQAssistant, Neo4j, Pandas, and Jupyter) for the analysis of data from different sources (such as JProfiler, Jenkins, and Git). Together, we have a look at how we can develop solutions to optimize performance, identify build breaker or make knowledge gaps in our source code visible.

Transcript

1. Software Analytics for Pragmatists Solving problems – automated, data-centric and reproducible Markus Harrer software analytics clean code
2. code
3. “Without data you're just another person with an opinion.“ W. Edwards Deming
4. Motivation Software Analytics “In software engineering there is much we are seeing, but little we are learning.” Tim Menzies
5. State of the Art
6. State of the Art
7. Questions from daily work
8. Motivation Why? •Make problems visible + Improve clarity and understanding •Drive decisions + Raise money + Raise more money for further analysis •Support Continuous Learning + Master challenges + Thrive for improvement steadily
9. Claim Solving problems – automated, data-centric and reproducible.
10. Claim Solving problems – automated, data-centric and reproducible.
11. Automation •Data retrieval •Analysis •Visualization DevOps style
12. Claim Solving problems – automated, data-centric and reproducible.
13. Types of software artifacts and meta data chronological community runtime static
14. Data is dirty Time for cleaning up data up to 80%
15. Claim Solving problems – automated, data-centric and reproducible.
16. Data + Automation + Traceability Replication
17. Code Book Example: Analyzing performance bottlenecks
18. Claim Solving problems – automated, data-centric and reproducible.
19. Pipeline Data Mining • NumPy • scikit-learn • SciPy Visualization • matplotlib • plot.ly • Bokeh • python-pptx ...
20. Pipeline XML/Graph Tabellen matplotlib Pandas, ... Pandas jQAssistant, Neo4j Text xlsx E pptx P Python Jupyter Input Pre- processing Analysis Output D3
21. Pipeline ZIP GZ *.class JAR, WAR, EAR MANIFEST.MF *.properties XSD YAML XML application.xml web.xml beans.xml JaCoCo FindBugs CheckStyle pom.xml surefire-reports.xml RDBMS Schema M2 Repository DB CSV Excel BigQuery Inputs HDFStore Web JSON Git Pandas jQAssistant
22. Examples Demos •Performance Bottlenecks •Build Breaker Analysis •Mining Knowledge Islands •Git Log Analysis
23. Experience Learning by Doing!
24. What‘s the value of the information related to the effort for the analysis of the information? Always ask the question
25. References Leek, Jeff: The Elements of Data Analytic Style. LeanPub, 2015. McKinney, Wes: Python For Data Analysis, O’Reilly, 2012. Mens, Tom; Serebrenik; Cleve, Anthony:Evolving Software Systems. Springer, 2014. Mens, Tom; Demeyer, Serge: Software Evolution. Springer, 2008. Shull, Forrest; Singer, Janice; Sjøberg, Dag I.K.: Guide to Advanced Empirical Software Engineering. Springer, 2008. Tornhill, Adam:Your Code As a Crime Scene. Pragmatic Programmers, 2015.
26. Your Questions