The Next 1100 Haskell Programmers

We report on our experience teaching a Haskell-based functional programming course to over 1100 students for two winter terms. The syllabus was organized around selected material from various sources. Throughout the terms, we emphasized correctness through QuickCheck tests and proofs by induction. The submission architecture was coupled with automatic testing, giving students the possibility to correct mistakes before the deadline. To motivate the students, we complemented the weekly assignments with an informal competition and gave away trophies in a award ceremony.

Joint work by Jasmin Blanchette, Lars Hupel, Tobias Nipkow, Lars Noschinsk, Dmitriy Traytel

Presented at the Haskell Symposium 2014, co-located with ICFP, Gothenburg, Sweden.

Transcript

1. The Next 1100 Haskell Programmers Jasmin Blanche e Lars Hupel Tobias Nipkow Lars Noschinski Dmitriy Traytel Technische Universität München September 4th, 2014
2. 1 Se ng & Syllabus 2 Exercises & Compe on 3 Exam
3. Curriculum Introduc on to Func onal Programming ▶ Compulsory course for 2nd year students ▶ Computer Science ▶ Informa on Systems ▶ Science Teaching ▶ Elec ve course for ▶ Mathema cs ▶ Economics ▶ ... 3 / 22
4. Why Haskell? Haskell has: ▶ A large user community ▶ Real-world appeal ▶ A variety of textbooks ▶ QuickCheck ... but also: ▶ Lazy evalua on 4 / 22
5. Course Outline 1. Introduc on to FP 2. Basic Haskell: QuickCheck, guarded equa ons, ... 3. List comprehensions, polymorphism, basic type classes, pa ern matching 4. Proof by structural induc on on lists 5. Higher-order func ons, λ-abstrac ons, extensionality 6. Type classes 7. Algebraic datatypes and structural induc on 8. I/O: files, web, ... 9. Modules: module syntax, data abstrac on, correctness proofs 10. Case study: Huffman coding 11. Lazy evalua on and infinite lists 12. Complexity and op miza on 13. Case study: parser combinators 5 / 22
6. Course Outline 1. Introduc on to FP 2. Basic Haskell: QuickCheck, guarded equa ons, ... 3. List comprehensions, polymorphism, basic type classes, pa ern matching 4. Proof by structural induc on on lists 5. Higher-order func ons, λ-abstrac ons, extensionality 6. Type classes 7. Algebraic datatypes and structural induc on 8. I/O: files, web, ... 9. Modules: module syntax, data abstrac on, correctness proofs 10. Case study: Huffman coding 11. Lazy evalua on and infinite lists 12. Complexity and op miza on 13. Case study: parser combinators 5 / 22
7. Course Outline 1. Introduc on to FP 2. Basic Haskell: QuickCheck, guarded equa ons, ... 3. List comprehensions, polymorphism, basic type classes, pa ern matching 4. Proof by structural induc on on lists 5. Higher-order func ons, λ-abstrac ons, extensionality 6. Type classes 7. Algebraic datatypes and structural induc on 8. I/O: files, web, ... 9. Modules: module syntax, data abstrac on, correctness proofs 10. Case study: Huffman coding 11. Lazy evalua on and infinite lists 12. Complexity and op miza on 13. Case study: parser combinators 5 / 22
8. Course Outline 1. Introduc on to FP 2. Basic Haskell: QuickCheck, guarded equa ons, ... 3. List comprehensions, polymorphism, basic type classes, pa ern matching 4. Proof by structural induc on on lists 5. Higher-order func ons, λ-abstrac ons, extensionality 6. Type classes 7. Algebraic datatypes and structural induc on 8. I/O: files, web, ... 9. Modules: module syntax, data abstrac on, correctness proofs 10. Case study: Huffman coding 11. Lazy evalua on and infinite lists 12. Complexity and op miza on 13. Case study: parser combinators Image credit: Randall Munroe 5 / 22
9. 1 Se ng & Syllabus 2 Exercises & Compe on 3 Exam
10. Weekly Assignments ... consist of three parts: 1. Group exercises 2. Homework 3. Compe on problem 7 / 22
11. Tutorial Groups ▶ 90 minutes ▶ Lead by a student TA ▶ ∼20–25 groups ▶ Up to 24 students per group ▶ Usually ∼10 show up 8 / 22
12. Homework ▶ 3–4 problems ▶ Time needed to solve problems greatly varies ▶ Students can upload their submissions via SSH ▶ Cronjob runs a test suite, mails results to students ▶ Students can re-upload arbitrarily o en before deadline ▶ Graded by student TAs 9 / 22
13. Test Suite ▶ Tests hand-wri en by us ▶ Checks against a specifica on and a reference implementa on ▶ Tests and expected results are secret, failed inputs are reported ▶ Problem: excep ons and non-termina on ▶ Output of within combinator poten ally confusing for students ▶ Excep ons captured with Control.Spoon ▶ Future work: automated style checks with HLint? 10 / 22
14. Test Suite Lessons Learned ▶ Ensure that tests are not vacuous (defaul ng to Unit!) ▶ Difficult to explain to students ▶ Instan ate polymorphic proper es with small types ▶ Write generators by hand, tuned to the problem at hand ▶ Tune parameters (maxSize, maxSuccess, ...) ▶ Do not trust “Detec ve QuickCheck” alone 11 / 22
15. Image credit: Roman Cheplyaka
16. Compe on 1. Some mes a dis nguished homework exercise, some mes op onal 2. Fix a ranking criterion ▶ Brevity ▶ Efficiency ▶ Ar s c value 3. Rank all student submissions 4. Comment and publish good solu ons 5. At the end of the year: Award ceremony 13 / 22
17. Compe on 1. Some mes a dis nguished homework exercise, some mes op onal 2. Fix a ranking criterion ▶ Brevity ▶ Efficiency ▶ Ar s c value 3. Rank all student submissions 4. Comment and publish good solu ons 5. At the end of the year: Award ceremony ... (mostly) run by Jasmin “MC Hammer” Blanche e 13 / 22
18. Compe on Mo va on The programming contests a ract a substan al frac on of the most talented students that we have ... It is our role to discover these talents and make them shine. – Piotr Rudnicki 14 / 22
19. Compe on Problem from week 2 Task: Write a func on [Char] -> [[Char]] returning all dis nct permuta ons of the input list in reverse alphabe c order. Criterion: Token count, then efficiency. 15 / 22
20. Compe on Problem from week 2 Task: Write a func on [Char] -> [[Char]] returning all dis nct permuta ons of the input list in reverse alphabe c order. Criterion: Token count, then efficiency. Winning entry perms chars = reverse $ sort [c : str | c <- nub chars, str <- perms $ delete c chars ] ‘max‘ [””] 15 / 22
21. Compe on Problem from weeks 11–12 Task: Write an AI player for Othello/Reversi Criterion: Ranking in a tournament between all contestants 16 / 22
22. Compe on Results ▶ Many students spent a lot of me on solving the problems ▶ ... some even ignored all the other exercises ▶ Ranking the solu ons takes a lot of me ▶ Vehicle to teach advanced topics outside of the syllabus ▶ Fosters library explora on ▶ Rewarding for us: recrui ng students 17 / 22
23. Compe on Results ▶ Many students spent a lot of me on solving the problems ▶ ... some even ignored all the other exercises ▶ Ranking the solu ons takes a lot of me ▶ Vehicle to teach advanced topics outside of the syllabus ▶ Fosters library explora on ▶ Rewarding for us: recrui ng students ▶ CYP – a simple proof checker (Durner) ▶ Primi ve corecursion for Isabelle (Panny) ▶ Waldmeister for Sledgehammer (Steckermaier) ▶ Complexity analysis of a small func onal language (Wimmer) ▶ Func onal congruence closure (Franze) ▶ ... and >5 future TAs 17 / 22
24. 1 Se ng & Syllabus 2 Exercises & Compe on 3 Exam
25. Exam Results in 2012 0 50 100 150 200 250 0 10 20 30 40 50 Homework points Examina on points Linear regression Passing score 19 / 22
26. Exam Results in 2013 0 50 100 150 200 250 0 10 20 30 40 Homework points Examina on points Linear regression Passing score 20 / 22
27. Exam What works well? ▶ Induc on and equa onal reasoning ▶ List comprehensions ▶ I/O What does not work so well? ▶ Type inference ▶ Higher-order func ons ▶ Evalua on order 21 / 22
28. Q & A