Opened 10 years ago

Last modified 8 years ago

#13555 new defect

Doctests of random functions needs to be improved.

Reported by: JoalHeagney Owned by: mvngu
Priority: major Milestone: sage-6.4
Component: doctest coverage Keywords: random, matrices, random_matrix, properties of random objects, properties
Cc: Merged in:
Authors: Reviewers:
Report Upstream: N/A Work issues:
Branch: Commit:
Dependencies: Stopgaps:

Status badges


Currently a lot of random functions are skipped during doc-testing, as their results can't be predicted.

However, in ticket #13554, it is clear that errors are slipping through, and properties of random objects should be checked when possible.

For example, in ticket #13554, the generated matrices should at least be tested for the following (if they aren't already):

  • Presence of zero entries when method/documentation states that no zero entries should be created, unless density= keyword is used.
  • method=echelon_form produces matrices in echelon form.
  • method=echelonizable can be echelonized.
  • method=unimodular has a determinant of 1.
  • method=diagonizable can be diagonalized and eigenvalues are integers.
  • x= , y= has all entries between x (inclusive) and y (non-inclusive), unless density= keyword is used and 0 isn't between x and y. (In that case entries should be either between x and y, or zero.)

Change History (12)

comment:1 Changed 10 years ago by JoalHeagney

Sorry, on the last dot point from the documentation, it should check for the following instead:

  • all entries e between x <= e < y, not including 0 even if 0 is in this range, unless density= keyword is used.

comment:2 Changed 10 years ago by rbeezer

I think some of the desired tests are performed on the subsidiary methods. There was a desire at the time not to put too much in the main documentation.

Fir example, random_unimodular_matrix tests three very different results for determinant one.

comment:3 Changed 10 years ago by JoalHeagney

But if there had been some, the inconsistency between documentation and result would have been spotted a long time ago. The bug that inspired this trac report has been around since 4.7.3. #11968

Version 0, edited 10 years ago by JoalHeagney (next)

comment:4 follow-up: Changed 10 years ago by JoalHeagney

This is some code I put together to check matrix values element by element:

A = random_matrix(ZZ,5); A
def checkfunc(matrix, func):
    for val in matrix.list():
        if func(val) == True:
            return True
    return False
checkfunc(A, lambda x: x == 0)

A = random_matrix(ZZ,5,x=4,y=10)
checkfunc(A, lambda x: (x >= 4 & x < 10))

It's been a while since I was an efficient python programmer, so I'm sure someone will show me a generator/list method which is a lot more efficient.

I did checkfunc as a function because I was thinking about large matricies and saving memory space (as well as returning on the first match to the conditions). I'm not satisfied with the fact that checkfunc iterates over a list rather than a generator.

comment:5 in reply to: ↑ 4 Changed 10 years ago by jhpalmieri

You might consider using all or any:

A = random_matrix(ZZ,5); A
def checkfunc(matrix, func):
    return any(func(val)==True for val in matrix.list()) # untested

Also, I'm not sure that there is an iterator for the elements of a matrix. I guess there's one for the rows (mat.__iter__()), and then I guess for each row there's one (row.iteritems()). I don't see one for all of the elements, though. (I tried searching the files in the matrix directory for "yield" and didn't find much.)

comment:6 Changed 10 years ago by JoalHeagney

I also got the second lambda function incorrect. We're testing if any values are outside the range 4 (inclusive) to 10 exclusive in my example matrix.

so it should be something like:

lambda x: (x < 4 | x >= 10)

or even better

lambda x: x not in range(4,10)

I did get an iterator-only method for getting elements of a matrix using this:

(A[valrow][valcolm] for valrow in xrange(A.dimensions()[0]) for valcolm in xrange(A.dimensions()[1]))

But then I realized that since we're only using this function (theoretically) in doc-checking, we probably don't have to worry about memory concerns and iterators versus lists. jhpalmieri's approach seems nice and simple.

comment:7 Changed 10 years ago by JoalHeagney

Theoretically we could use jhpalmieri's approach thus:

A = random_matrix(ZZ,5); A
any((lambda x: x == 0)(val)==True for val in A.list())

to completely avoid defining checkfunc at all. Don't know if this would be an advantage or not during doc-checking.

comment:8 Changed 10 years ago by JoalHeagney

Found a way to iterate over elements in a matrix using a generator.

if M is a matrix, then this:

(M[vals[0],vals[1]] for vals in xmrange(M.dimensions()))

will return a generator object that does the job.

comment:9 Changed 9 years ago by jdemeyer

  • Milestone changed from sage-5.11 to sage-5.12

comment:10 Changed 8 years ago by vbraun_spam

  • Milestone changed from sage-6.1 to sage-6.2

comment:11 Changed 8 years ago by vbraun_spam

  • Milestone changed from sage-6.2 to sage-6.3

comment:12 Changed 8 years ago by vbraun_spam

  • Milestone changed from sage-6.3 to sage-6.4
Note: See TracTickets for help on using tickets.