[SciPy-user] numarray complex comparisons

[Prabhu Ramachandran]

Hi,

>>>>> "EJ" == eric jones <eric at enthought.com> writes:

    EJ> Sigh.  I am weary of this topic.  I've discussed it on
    EJ> multiple occasions with many people -- both by email and
    EJ> face-to-face.  Some people don't care -- give them a rule and
    EJ> they'll adapt.  The remainder is about 50% split.  I guess

Sure, I agree, folks will get used to whatever assumptions that are
made.

    EJ> this mirrors other numeric computing communities.  Some
    EJ> Fortran compilers use one approach, some use the other.
    EJ> Matlab uses one approach, IDL uses the other.  Seems like
    EJ> people either argue from the "moral high ground" (as Prabhu
    EJ> does here) that it just ain't right to compare complex values,
    EJ> or they argue pragmatically (as Travis O. and I do) that it
    EJ> breaks generic array code that uses comparisons.

Its not really a "moral high ground".  I was trying to give an
alternative argument.  I've only seen "pragmatic" views this far so
thought a different point of view would be good.  Also, I was trying
to make a generic design/philosophy argument.

    EJ> I'm happy for others to discuss it, but I doubt I spend any
    EJ> more time on it after this.  My experience has been that this
    EJ> is a "Mac is better than Windows" argument.  Everyone has a
    EJ> nice argument and then goes home with the same opinion and not
    EJ> believing how stupid the other side was not to "get it."

Well, just to get it straight I never said anyone was stupid in my
email and I did not even venture to think that.  In fact I said what I
did because we were asked.  What I am comfortable with or not does not
make others who disagree stupid and what I am comfortable may well
change in time.

[snip]

    EJ> By the way, the choice to use real is useful in many cases.
    EJ> There are algorithms that return complex arrays with very
    EJ> small imaginary parts do to round-off error.  In this case
    EJ> ignoring the imaginary part is often what you want to do.

Ok, so this is a democratic decision then and you seem to have
evidence that this is the case.  Fine.

[snip]

    >>  So, I think that you should simply raise an exception if you
    >> expected a real number and got a complex number instead.  If
    >> this is way too painful to have to deal with then make some
    >> assumption and print it clearly each time you make that
    >> assumption so that the user knows exactly what is going on and
    >> where.  I think this is the cleanest solution that I'll be
    >> confortable with.  Anything else makes me uncomfortable.  Maybe
    >> its because I am not used to this, I dont know.

    EJ> Printing it out is an option -- although I'm not sure I like
    EJ> it because it might become *very* noisy in some cases.  This
    EJ> is worth some thought.

OK, then atleast something useful came out of my efforts. :)

    EJ> I don't really expect this to convince anyone that comparing
    EJ> reals is better.  It merely states my rationale and gives the
    EJ> precedence for comparing reals.  Those fearful of all out
    EJ> calamity should take solace in the fact that the most used
    EJ> numeric scripting language on the planet uses the same
    EJ> approach and has proven to be very useful indeed.

My concern was simple.  By making an assumption on how to order
complex numbers will we affect results of someones computations?  I
dont know and was thinking of a worst case scenario (which if you ask
me is not a bad thing to do).  I cant argue with Matlab's
experience. :)

cheers,
prabhu