Agenda

Admin (10min)

Discuss readings

General: Sumit Gulwani’s Spreadsheet Data Manipulation Using Examples (in email)

Suggested Background: Version Space Algebra and its Application to Programming by Demonstration 

Additional reading: Martin Rinard’s Example Driven Program Synthesis for End-User Programming

Possible followup with Gulwani: Feedback generation for intelligent tutoring systems?

The Tozier files

Pushing Zeke to Push and Clojure

“Fast” lexicase selection

Quotes re GECCO?

Finite algebra draft

Clojush updates needing merging? (recognize-literal)

Meeting Notes

• Lee will be meeting with Hampshire PR person about GECCO article. Get him quotes about GECCO if you want!
• We have merged the changes to master. NOTE: Make sure you follow the proceedure for updating code – don’t update on GitHub, since things get messed up that way.
• Finite Algebras paper draft will be emailed to those interested in reading and making suggestions.
• Tom will be using the cluster next week for some kata bowling runs.
• Zeke was encouraged to use Clojure and Push, but will likely stick to what he knows.
• Tom will delete fast-lexicase-selection and push to GitHub. The old version will still be on GitHub in old commits, so we can get it back later if we want.

The Tozier Files

• There are some valuable things in the Bill Tozier emails, but also some things that aren’t really what we want to do.
• Things that may be valuable to persue:
  • Try shorter test cases with kata bowling
  • Monitoring of runs during evolution. Piano Roll plots or other finer details may be worthwhile, but other graphical investigations that only require things that are currently printed in logs may be just as good.
  • Think about why we are failing, and what to try next.
  • Cheating: Interesting topic. It may be useful to cheat at the beginning to see what you are up against, and take away later once you understand the problem better. Plus, cheating can be seen in different lights depending on your goals.
  • Maybe remove some instructions to make the search space smaller (flush, yank, shove, …)
  • Try writing a solution by hand.

Discussion of Gulwani paper

• This makes us think it may be worthwhile to implement a stronger string / pattern matching library in Push.
• The problems solved here may be inspirations for future “programming problems” for GP to solve.
• Our motivations for program synthesis are similar to theirs.
• As for the intelligent tutoring system, Lee will contact Sumit Gulwani to investigate further collaboration.