CODE ANALYSES USING
NATURAL LANGUAGE
QUESTIONS
Software Composition Seminar
Student: Michael Zbinden
Supervisor: Pooja Rani
29. May 2018
Problem
• Huge code base
• Finding classes
• Example: What are the deprecated classes?
2
Tools Support
• IntelliJ Search
3
Tools Support
• Eclipse Search
4
Moose
5
Visualize code in Moose
6
Software Representation
7
Challenges
• Learn about tool
• Know the meta-model
• Learn query language
• Barrier
8
Solution
• Smalltalk Code
• Use Natural Language
9
Generate Code
• Python experiment
10
About Paper
• Translation problem
• Accuracy
• Neural network model
• Model adoption
11
Tools used
• Framework: Theano
• Library: NLTK
• AST: astor
• Language: Python
12
Paper
13
Challenges in Running Model
• Complex neural network
• Lots of components
• Running Code
• Dataset available for python
14
Findings
• Difficult to prepare dataset
• Creating neural network
• Neural Network libraries
• NLP library
15
Learning challenges
• Learning about moose
• Creating a simple query
• Analyzing complex neural network
• Python Dependency
16
Future work
• Prepare dataset
• Decoupling Code
• Adapt classes for other datasets
17