On Mon, Feb 9, 2015 at 12:40 PM, Ondřej Čertík <[email protected]> wrote:
> Hi Peter,
>
> On Mon, Feb 9, 2015 at 9:09 AM, Peter Chervenski <[email protected]> wrote:
>> I made this function to test for the equivalence of two expressions. It
>> doesn't really prove anything, but if the tests are many, the probability of
>> it being wrong becomes negligible. Do such utility functions have a place in
>> SymPy?
>>
>> def equiv(a, b, ntests=15):
>>     """ Test if expression a is equivalent to b
>>     by comparing the results of many random numeric tests """
>>
>>     # get the symbols
>>     sb_a = filter(lambda x: x.is_Symbol, a.atoms())
>>     sb_b = filter(lambda x: x.is_Symbol, b.atoms())
>>
>>     sb = list(set(sb_a + sb_b))
>>
>>     eq = True
>>     for i in xrange(ntests):
>>         k = dict(zip(sb, np.random.randn(len(sb))))
>>
>>         r_a = a.subs( k )
>>         r_b = b.subs( k )
>>
>>         # prove there is a difference
>>         if (r_a - r_b)**2 > 1e-30: # not the same? the expressions are
>> different
>>             eq = False
>>             break
>>
>>     return eq
>
>
> Absolutely. I've also implemented a similar function in one PR:
>
> https://github.com/sympy/sympy/pull/8036/files?diff=unified#diff-2c9ef1ef2c82f5d5781d0d12e1fe4910R33
>
> It was pointed out to me that we have similar machinery here already:
>
> https://github.com/sympy/sympy/blob/master/sympy/utilities/randtest.py#L43
>
> This should be unified and put into sympy. We can call it
> "zero_numerical" or something like that ("test_numerically", ...).
> Mathematica calls this PossibleZeroQ (though I think it does both
> symbolic an numerical tests).
>
> Look at the implementation in my PR --- you should allow the user to
> specify the range (I call it [a, b]) as well as the precision. I think
> we can perhaps just test for 0, and then your "equiv" can just call
> this zero testing function for a-b.

Actually, in my implementation I choose random integers from [-a, a]
and then test an interval
[-a/b, a/b]. That way you will get rational numbers, as opposed to
only integers (that could hide differences).

Ondrej

-- 
You received this message because you are subscribed to the Google Groups 
"sympy" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/sympy.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/sympy/CADDwiVDnybsk6Ke-ocGoPd%2BuB_S7v%3DDCeSgvGTR1JFTXiw5F_g%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to