Re: [Open Babel] Loss of precision when copying pybel.Molecule

[Noel O'B. <bao...@gm...>]

(ccing to list)

On 24 March 2013 21:23, inviso <in...@gm...> wrote:
> 1) Good deal, looks like that was what I was seeing
>
> 2) Yeah, I figured that out a few minutes after sending my message but
> hadn't see it show up on the list so I could reply.  The things you remember
> to check after hitting send... :(
>
> 3) Yep, that's what I ended up with and it's doing a proper deep copy as far
> as I can tell and is preserving values as expected.
>
> 4) Thanks.  I'm loading up larger files typically.  Any reason not to load
> up a set of molecules the way I was loading in the one test molecule?  aka,
> molecules = [mol for mol in pybel.readfile...].  I want them in a list
> either way.

Regarding (4), there is no particular reason not to, but if you ever
find yourself using "z = [x for x in y]" then a better Python idiom is
"z = list(y)". The intention is then somewhat clearer.

- Noel

> Greg
>
>
> On Sun, Mar 24, 2013 at 1:17 PM, Noel O'Boyle <bao...@gm...> wrote:
>>
>> Hi Greg,
>>
>> 1. "mol_from_Molecule_copy = pybel.Molecule(mol)"
>> This doesn't copy anything. It writes it out as a Mol file, and then
>> reads it in from a Mol file. A Mol file has 4 decimal places of
>> precision. It seems that I should document this.
>>
>> 2. "mol_from_OBMol_copy = pybel.Molecule(mol.OBMol)"
>> This doesn't copy anything either. It just creates a pybel Molecule as
>> a thin wrapper around an OBMol.
>>
>> 3. If you want to copy an OBMol, you should use OBMol's copy
>> constructor, "myCopyMol = pybel.ob.OBMol(mol.OBMol)"
>>
>> 4. FYI, "mol = [mol for mol in pybel.readfile('xyz', 'test1.xyz')][0]"
>> is better as "mol = pybel.readfile(...).next()"
>>
>> HTH,
>> - Noel
>>
>> On 24 March 2013 18:36, inviso <in...@gm...> wrote:
>> > In one specific case that I'm able to identify, I'm losing precision on
>> > the
>> > geometry vectors when copying a molecule in pybel.  I've attached a
>> > small
>> > sample of source and a single molecule file that reliably reproduces the
>> > results for me on Windows with OpenBabel 2.3.1 and Python 2.7.2.  Also,
>> > confirmed on Rocks Linux with OpenBabel 2.3.2 and Python 2.6.6.
>> >
>> > The output on my machine from the sample code is as follows:
>> > Original       : 0.0602900000, -2.2281500000, 0.0378600000
>> > Molecule_copy  : 0.0603000000, -2.2281000000, 0.0379000000
>> > OBMol_copy     : 0.0602900000, -2.2281500000, 0.0378600000
>> >
>> > A few notes:
>> > 1) Yes, the precision in my print format string is ridiculous :)  I just
>> > wanted to make sure I wasn't arbitrarily clipping precision when
>> > displaying
>> > the output.
>> > 2) The sample code only outputs the first atom (Rh) as it was all that
>> > was
>> > necessary to show the problem on my machine.
>> > 3) Since this works fine when passing OBMol into the constructor, it
>> > looks
>> > like it has to be caused by the Molecule._exchange call which in turn
>> > calls
>> > Molecule.write(format='mol')
>> > 4) Calling Molecule.write(format='mol') directly also shows a loss of
>> > precision on the geometry in the output.  The .write() call goes into
>> > openbabel.OBConversion and from there into swig code which I'm not very
>> > familiar with so I didn't try to follow it deeper into the C++ library.
>> > 5) It seems like the precision being lost is far greater than I would
>> > expected from the inherent precision problems of double and it feels
>> > more
>> > like it's just rounding to four digits of precision.  It seems
>> > especially
>> > suspect since using OBMol is accurate.
>> >
>> > I will work around this by always using new_copy =
>> > pybel.Molecule(my_mol.OBMol), but it is an odd behavior.  Is anyone else
>> > able to confirm?  Is it unexpected behavior to pass pybel.Molecule
>> > instead
>> > of OBMol or is the _cinfony check in the constructor specifically meant
>> > for
>> > that case?
>> >
>> > Thanks,
>> >
>> > Greg Fortune
>> >
>> >
>> > ------------------------------------------------------------------------------
>> > Everyone hates slow websites. So do we.
>> > Make your web apps faster with AppDynamics
>> > Download AppDynamics Lite for free today:
>> > http://p.sf.net/sfu/appdyn_d2d_mar
>> > _______________________________________________
>> > OpenBabel-discuss mailing list
>> > Ope...@li...
>> > https://lists.sourceforge.net/lists/listinfo/openbabel-discuss
>> >
>
>