Upload
ivano-malavolta
View
139
Download
2
Embed Size (px)
Citation preview
‹#› Het begint met een idee
Experiment reporting
Ivano Malavolta
Vrije Universiteit Amsterdam
2
Roadmap
● Experiment reporting
● Examples of empirical studies in SE
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
3
Let’s take a step back
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
19 Ivano Malavolta / S2 group / Empirical software engineering
5 - Presentation & package
● Document the results
● Prepare replication package
● Sum up lessons learned
● Write down reflectionsExperiment
scoping
Experiment planning
Idea
Experiment operation
Analysis & interpretation
Presentation & package
Vrije Universiteit Amsterdam
4 Ivano Malavolta / S2 group / Experiment reporting and recommendations
Basic structure of an experiment report
(Structured) Abstract
Introduction/Motivation
(Related Work)
Experiment design
Experiment execution
Data analysis
Threats to validity
Discussion
Conclusions
References
Vrije Universiteit Amsterdam
5
Abstract
● People judge papers by their abstracts○ usually decide whether to read the whole paper based on abstract
● It's important for the abstract to tell the whole story○ background and context
○ goals
○ method
○ results
○ conclusion
● Often it is the only part freely accessible
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
6
Mini-lab
1. Read the 3 abstracts
2. Rank them according to how they provide insights about:
a. problem being addressed
b. obtained results and main contributions
c. type of evaluation/empirical method
3. Which abstract is easier for a secondary study?
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
7
The structured abstract
● Context
○ brief explanation of the motivation for conducting the study
● Objective
○ aim, objects, focus, perspective of the study (rephrase GQM)
● Method
○ experimental design, number and kind of objects/subjects, selection
criteria, data collection and analysis procedures
● Results
○ main findings
● Conclusion
○ impact of the obtained results
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
8
Example of structured abstract
Vrije Universiteit Amsterdam
9
Introduction and motivation
● Introduction
○ Mini-version of the whole paper
■ Attacked problem■ Proposed solution or experiment■ Main results■ Main contributions (how to “use the paper”)■ Structure of the paper
● Motivation:
○ Scope of the work○ Problem statement○ Research objectives○ encourages to read the paper
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
10
Related work
● Why this paper exists?
● Complete picture about:
○ experiments with similar goals○ experiments with similar objects/subjects○ papers with similar goals, but not empirical → they lack in provided
evidence
● Two main strategies:
○ Paper-by-paper comparison○ Catalogue of related papers + overall comparison
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
11
Experiment design and execution
● Design: rewalk through all the experiment plan
○ GQM○ hypotheses○ objects and subjects○ variables ○ measurements○ experiment design○ data analysis strategies (e.g., used statistical tests)
● Execution: how the experiment plan has been put in practice
○ preparation (eg code instrumentation)○ data collection procedure○ data clean up○ any deviation from the plan○ reference to replication package
Vrije Universiteit Amsterdam
12
Analysis and threats to validity
Ivano Malavolta / S2 group / Experiment reporting and recommendations
● Analysis (or Results)
○ Demographics○ Descriptive statistics and data exploration○ For each research question:■ hypothesis testing■ discussion of effect size■ quick elaboration on obtained results
● Threats to validity○ Internal○ External○ Construct○ Conclusion
Vrije Universiteit Amsterdam
13
Discussion and Conclusions
Ivano Malavolta / S2 group / Experiment reporting and recommendations
● Discussion
○ Elaborate on the obtained results○ Impact into practice (industry)○ Make the results actionable○ Lessons learnt
● Conclusions○ Summary of the main contributions of the paper○ Future work■ other experiments on different aspects■ better ways to perform the experiment■ ...
Vrije Universiteit Amsterdam
14 Ivano Malavolta / S2 group / Experiment reporting and recommendations
More details? You have a full paper about experiment reporting!
Chapter 8
Vrije Universiteit Amsterdam
15
Replication package
3. Figures
○ exploratory figures
○ final figures
You have to follow this structure when providing the replication package of your final assignment!
1. Data
○ raw data (usually CSV)
○ processed data (CSV)
○ (name it properly!)
2. Source code (commented!)
○ measurement scripts (Python)
○ data processing scripts (R)
○ analysis scripts (R)
4. Text
○ README file
○ (optional) experiment notes
Vrije Universiteit Amsterdam
16
Examples of empirical studies in SE
● Nejati, Javad, and Aruna Balasubramanian. "An in-depth study of mobile browserperformance." Proceedings of the 25th International Conference on World WideWeb. International World Wide Web Conferences Steering Committee, 2016.
● Hassan, Safwat, Weiyi Shang, and Ahmed E. Hassan. "An empirical study ofemergency updates for top android mobile apps." Empirical Software Engineering(2016): 1-42.
● Tufano, Michele, et al. "An empirical investigation into the nature of test smells."Proceedings of the 31st IEEE/ACM International Conference on AutomatedSoftware Engineering. ACM, 2016.
● Linares-Vásquez, Mario, et al. "Mining energy-greedy api usage patterns in androidapps: an empirical study." Proceedings of the 11th Working Conference on MiningSoftware Repositories. ACM, 2014.
● Procaccianti, Giuseppe, Héctor Fernández, and Patricia Lago. "Empiricalevaluation of two best practices for energy-efficient software development."Journal of Systems and Software 117 (2016): 185-198.
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
17
What this lecture means to you?
● You know how to present your experiment
○ push for clarity, soundness, and completeness
● Replication package needed
○ helps community building
○ enables replications
○ gives you also visibility!
● Examples of very good papers
○ Be inspired!
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
18
Readings
Chapter 11 Chapter 16 Chapter 8
Ivano Malavolta / S2 group / Experiment reporting and recommendations
Vrije Universiteit Amsterdam
Next lecture
Guest lecture by Fabio Palomba
○ VU Green Office http://www.greenofficevu.nl/
○ Room HG KC-18
○ Thursday October 19th, 11 AM
TOPIC: Information retrieval for code changes based on user
reviews of mobile apps
Vrije Universiteit Amsterdam
Next steps
Vrije Universiteit Amsterdam8
Courseschedule
3 types of lectures:
• Theory
• Tutorial
• Hands-on session
Vrije Universiteit Amsterdam14
Scheduleanddeliverables
• Deliverable 1 (20% of the final grade)
• Experiment goal and scope description
• Deliverable:
• written report
• Deadline: 21 September: 23:59
• Deliverable 2 (30% of the final grade)
• Detailed design of the experiment
• Deliverable:
• written report
• Deadline: 5 October: 23:59
• Deliverable 3 (50% of the final grade)
• Final report of the experiment
• Deliverable:
• written report
• experiment execution scripts and source code
• raw data
• analysis scripts in R
• Deadline: 29 October: 23:59