On 29/11/2011 20:56, Noel O'Boyle wrote:
> On 2 November 2011 22:35, Chris Morley<[email protected]>  wrote:
>> On 01/11/2011 02:27, Geoff Hutchison wrote:
>>>
>>> On Oct 11, 2010, at 4:18 PM, Chris Morley wrote:
>>>
>>>> A simpler and more obvious model for this purpose has essentially a single 
>>>> IMPVAL for each charge state of the molecule. (Only if you are interested 
>>>> in radicals or hydrogen on the higher valency states of second row 
>>>> elements do you need another rule for each higher valence.) There is no 
>>>> need for any skilful fine tuning. It is more maintainable and will be 
>>>> faster because not so many SMARTS patterns need to be matched. Up to now, 
>>>> It has worked for everything I've tried (although this not very 
>>>> extensive), except with test_formula, where the fault is in a couple of 
>>>> erroneous results in formularesults.txt, which at least shows the old 
>>>> model was error-prone and needs some more tweaking of phosphate structures.
>>>
>>> I'd like to re-visit this idea now that we have some flexibility in the 
>>> development trunk. I tried last-year's patch, and saw ~14% speedup when 
>>> computing 279,000 formulas from a SD file. (The improvement comes from 
>>> eliminating the hybridization assignment.)
>>>
>>> (It also has the side benefit that we can try out ideas for MolCore rules 
>>> while testing them against existing OB tests.)
>>>
>>> Thoughts?
>>
>> I think we need a 2.3.2 release fairly soon. There are a number of bugs,
>> some of which were new in 2.3.1 which we need to correct. Could we aim
>> for the beginning of January? There needs to be a tag copy of 2.3.1 and
>> then I guess we can use the 2-3-x branch for these corrections, some of
>> which are already in trunk. Will you make the tag and do the merging?
>>
>> We can then use the trunk to try the new valence model and your success
>> with it is encouraging. I could upload it, unless you want to, when you
>> have done the merging. I realize that it is not essential to wait, but
>> it avoids possible complications. Since the trunk would then be aimed at
>> 2.4.0 I guess it then would be appropriate to add mods which destroy the
>> binary compatibility? (In the fingerprint code particularly, when I get
>> round to it.)
>
> To resurrect this thread...I think we should just go with it. It's
> possible to tag "after the fact" (see bottom of
> http://svnbook.red-bean.com/en/1.0/re07.html) so let that not stop us.
>
> Anyway, I've finally gotten around to testing the code (sorry!). If
> the code is merged, I can describe a test case which I think can be
> used to flush them out. Currently, it looks twice as good as the
> current code (if I've interpreted the results correctly) but there's a
> bit of work to be done still.

OK, lets do it. But I'm away and can't contribute until next Tuesday.

Chris


------------------------------------------------------------------------------
All the data continuously generated in your IT infrastructure 
contains a definitive record of customers, application performance, 
security threats, fraudulent activity, and more. Splunk takes this 
data and makes sense of it. IT sense. And common sense.
http://p.sf.net/sfu/splunk-novd2d
_______________________________________________
OpenBabel-Devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openbabel-devel

Reply via email to