1. Structured programming 4 Day 34 LING 681.02 Computational Linguistics Harry Howard Tulane University

2. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University2 Course organization  http://www.tulane.edu/~ling/NLP/ http://www.tulane.edu/~ling/NLP/

3. Structured programming NLPP §4

4. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University4 Today's topics  Defensive programming  Debugging  Algorithm design

5. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University5 Defensive programming  Brainstorm with pseudo-code  Careful naming conventions  Bottom-up construction  Functional decomposition  Comment, comment, comment  Regression testing

6. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University6 Brainstorm with pseudo- code  Before you write the first line of Python code, write what your program does as pseudocode.  That is to say, before writing a program that NLTK understands, write it in a way that people understand.

7. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University7 An example of pseudo- code  SPOT, move forward about 10 inches, turn left 90 degrees, and start moving forward, then start looking for a black object with your ultrasonic sensor, because I want you to stop when you find a black object, then turn right 90 degrees, and move backward 2 feet, OK?  What is good or bad about this example

8. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University8 A different phrasing of the example  SPOT, move forward about 10 inches and stop.  Now turn left 90 degrees.  Start moving forward, and turn on your ultrasonic sensor.  Stop when you find a black object.  Turn right 90 degrees and stop.  Move backward 2 feet and stop.  What is good or bad about this example?

9. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University9 Pseudo and real code  The main advantage of the second phrasing is that we can match up the commands in each line to elements in the programming language.

10. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University10 Careful naming conditions  Choose meaningful variable and function names.

11. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University11 Bottom-up construction  Instead of writing a 20-line program and then testing it,  build and test smaller units,  and then combine them.  In general, these smaller units should be functions.

12. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University12 NLP pipeline Fig. 3.1

13. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University13 Commenting  Add comments to every line,  unless what a line is does is so obvious that a comment would get in the way.  Your pseudo-code could become the comments on your real code.

14. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University14 Regressive testing  Keep a suite of test cases.  As your program gets bigger, it should still work on previous test cases.  If it stops working, it has 'regressed'.  A change in code has the (unintended) side effect of breaking something that used to work.  doctest module does testing  It runs a program as if it were in interactive mode.  See doctest documentation.

15. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University15 Debugging topics  Check your assumptions  Exception > stack trace  Interactive debugging  Python's debugger  Prediction

16. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University16 Debugging  "Most code errors result from the programmer making incorrect assumptions". (NLPP:158)  When you find an error, first check your assumptions.  Add print statements to show  values of variables and  how far the program progresses.  Reduce input to smallest amount needed to cause the error.

17. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University17 Stack trace  A runtime error (Python exception) gives a stack trace that pinpoints the location of program execution at the time of the error.  But the error may actually be upstream.

18. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University18 Python's debugger  Invoke it:  import pdb  pdb.run('mymodule')  It lets you  monitor execution of program,  specify line numbers where program should stop (breakpoints), and  step through the sections of code inspecting values of variables.

19. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University19 Prediction  Try to predict the effect of a potential bugfix before re-running the program.  "If the bug isn't fixed, don't fall into the trap of blindly changing the code in the hope that it will magically start working again." (NLPP:159)  For each change, try to articulate what is wrong and how the change will fix the problem.  Undo the change if it doesn't work.  "Programs don't magically work; they magically don't work." (Robert Goldman)

20. Algorithm design NLPP 4.7

21. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University21 Algorithms  Divide and conquer  Start with something that works  Iteration  Recursion

22. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University22 Divide and conquer  Divide a problem of size n into two problems of size n/2.  Binary search - dictionary example.

23. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University23 Start with known  Transform task into something that already works.  To find duplicates in a list,  first sort the list,  then check for identity of adjacent pairs.

24. 16-Nov-2009LING 681.02, Prof. Howard, Tulane University24 Iteration vs. recursion  For some function ƒ…  Iteration  Repeat ƒ some number of times.  Calling ƒ in a for loop.  Recursion  ƒ calls itself some number of times:  NP → the N PP.  PP → P NP.

25. Next time Start NLPP §6 Learning to classify text