Skip to content

Worksheet: Jonathan

Jan Svačina edited this page Oct 25, 2019 · 6 revisions

Iteration II.

Worksheet

Week Hours Issue Description Evidence
Week 6. 7.5 SCA-10 Initial theorizing of idea, brainstorming, collaboration with Jan and investigation of ML techniques. Created Python code to randomly create test data out of a dictionary of line types. presentation
Week 7. 8 SCA-20 Was investigating methods of interpreting LoC as types of vectors, searching for related work material, preliminary experiment with naive classification types
Week 8. 10 SCA-11, SCA-6, SCA-13, SCA-14 Proposing classification model, experimenting with neural network configurations, creating prototype, Generating random data for training instead of dictionary now with a slightly smarter predicate generator so the structure is predictable. 8ce266c, f09990d
SUM: 25.5

Kanban Tasks

Code Name Description Asigned To Status
SCA-10 Creating Randomized Java Line Segments for Testing We need to be able to test our model, and hand classifying thousands and thousands of lines of code is unreasonable. Jonathan Backlog
SCA-17 Create tasks and keep up with them for each classification type We need a task to track progress for every type of classification in order to create random data for training. We have 2, loop and conditional. We need many more tasks. This would add about 10 tasks at minimum. Jonathan Backlog
SCA-19 Investigate methods of interpreting virtual LoC Word associations, vectorization, corpus all vs. tokenizing. Tokenizing is not viable, we need meaning of each token or piece of the code. Jonathan Backlog
SCA-18 Research Related Work Produce a list of related work papers that we can use as references both for research and paper writing. Jonathan In Development
SCA-13 SCA-10: Loop Statements Create a program that can create expected input to the model, namely, loop statements (do, for, while, each). Jonathan Review
SCA-14 SCA-10: Conditional Statements Same as SCA-13 but for conditionals. Jonathan Review
SCA-15 Prototype Machine Learning Model We have a prototype ML model that can determine with up to 84% accuracy between a line of Java code that is conditional vs. loop. Jonathan Test

Contribution Statistics

  • 3 git commits
  • numerous research papers discovered and read
  • many hours of learning how to use keras/tensorflow
  • many hours brainstorming with Jan

Clone this wiki locally