Re: [Joelib-help] RE: joelib.test.DescriptorSelection

[Joerg K. W. <we...@in...>]

Hello,

> Then I edited the joelib.properties stripping out all the "jcompchem.".
That's perfect !!! I'm sorry for that problems again ...
> Then I built all thing (with several messages). I replaced the joelib.jar of the release 20030625 with the new one.
> I tried reading the source, but I could not guess the format of the "descNamesURL".
o.k. that's definitely a bug:

Here is the bug fix and the descSelection scripts to facilitate the process:
descSelection.bat
descSelection.sh
src/joelib.properties
src/wsi/ra/tool/ResourceLoader.java

Then i used:
cygwin>sh descSelection.sh -iSDF 
d:/workingAt/datasets/algorithmTuning/regression2/moe/LogPS/huuskonen.sdf 
-oSDF test.sdf 
d:/workingAt/datasets/algorithmTuning/regression2/moe/LogPS/huuskonen.inputs 
normal " "

Problem description:
The resource loader is really important, because it can load resource 
files from files AND archives, likr .jar/.zip.
The problem was, that it has tried to load c:/yourDir as URL which does 
only work if you use file:/c:/yourDir

I've fixed it and it works noe also with:
X:/yourdir
and
/cygdrive/x/yourDir

By the way you can change src/log4j.properties for debugging purpose, e.g.
change:
#WSI
log4j.category.wsi.ra.io.BatchScriptReplacer=DEBUG
log4j.category.wsi.ra.tool.ResourceLoader=INFO
to:
#WSI
log4j.category.wsi.ra.io.BatchScriptReplacer=DEBUG
log4j.category.wsi.ra.tool.ResourceLoader=DEBUG

and you will see was the ResourceLoader is doing. I hope this will not 
necessary again !;-)

> Regards, Ambrogio

Regards, Joerg
> 
> -----Original Message-----
> From: Joerg K. Wegner [mailto:we...@in...]
> Sent: Wednesday, June 25, 2003 4:59 PM
> To: Oliva Ambrogio; JOELib help
> Subject: Re: joelib.test.DescriptorSelection
> 
> 
> Hello Ambrogio,
> 
> 
>>I modified the descNameFile as suggested but I keep receiving the same NullPointerException. Maybe I've put the wrong "delimiter" on the command line?
> 
> mmmh, oh sorry ... that's my fault, because i'm using another 
> joelib.properties file.
> 
> Please add the following lines to the src/joelib.properties file (after 
> joelib.filter.5.descriptionFile ...):
> 
> joelib.filter.6.name              =DescriptorFilter
> joelib.filter.6.representation    =joelib.process.filter.DescriptorFilter
> joelib.filter.6.descriptionFile   =docs/process/descriptorFilter
> 
> 
>>Thanks for the quick reply.
> 
> I'll hope this will work. I'll release a new JOELib version with the bug 
> fixes this day.
> 
> 
>>Regards, Ambrogio
> 
> Regards, Joerg
> 
> 
>>PS. Your PSs are a little bit too technical for my knowledge of java programming.
> 
> P.S.:This is possible !;-)
> 
> 
>>-----Original Message-----
>>From: Joerg K. Wegner [mailto:we...@in...]
>>Sent: Wednesday, June 25, 2003 3:46 PM
>>To: Oliva Ambrogio; JOELib help
>>Subject: Re: joelib.test.DescriptorSelection
>>
>>
>>Dear Dr. Ambrogio,
>>
>>
>>
>>>first of all thanks for making available JOELib.
>>
>>thanks ...
>>... i'm crossposting this message also to the JOELib-help mailing list.
>>You can join if you want ...
>>http://sourceforge.net/mail/?group_id=39708
>>
>>
>>
>>
>>>I successfully used the Convert application for adding descriptors to SD files. However I do not need all the descriptors. Then I tried to use DescriptorSelection to "select" the descripors I need.
>>
>>That's exactly the same way i use it, especially when calculating 
>>autocorrelation or other more extended descriptors (not public).
>>
>>
>>
>>>The problem is that it's not clear to me the syntax, in particular the format of the descriptors list and what the "delimiter" is.
>>
>>Let's try to resolve that...
>>
>>
>>
>>>I tried: 
>>>C:\FreeSoft\JOELib>java -cp %CLASSPATH% joelib.test.Convert -iSDF -oSDF +d c:\temp\sample.sdf c:\tem
>>>p\sample_wd.sdf
>>
>>All fine !
>>
>>
>>
>>>Then:
>>>C:\FreeSoft\JOELib>java.exe -cp %CLASSPATH% joelib.test.DescriptorSelection -iSDF c:\temp\sample_wd.
>>>sdf -oSDF c:\temp\sample_wd2.sdf c:\temp\Sample_desc.txt normal ;
>>>13:34:44 [INFO ] joelib.data.JOEElementTable              - Using element table: joelib/data/plain/e
>>>lement.txt
>>>13:34:44 [INFO ] joelib.io.IOTypeHolder                   - 12 input/output types loaded.
>>>13:34:44 [INFO ] joelib.process.filter.FilterFactory      - 5 filter informations loaded.
>>>Exception in thread "main" java.lang.NullPointerException
>>>       at joelib.test.DescriptorSelection.parseCommandLine(DescriptorSelection.java:167)
>>>       at joelib.test.DescriptorSelection.main(DescriptorSelection.java:301)
>>>
>>>where c:\temp\Sample_desc.txt is:
>>>
>>>Number_of_bonds;LogP;Topological_radius
>>
>>No, this should be simply:
>>
>>Number_of_bonds
>>LogP
>>Topological_radius
>>
>>so you can simply use the JOELib statistic-file and import it under 
>>Excel and extract the descriptor names you want !;-)
>>
>>
>>
>>>Can you help me in getting DescriptorSelection work?
>>>Thanks in advance.
>>
>>Was me a pleasure.
>>
>>
>>
>>>By the way I've also have a couple of "bug reports":
>>>- In the Tutorial I've noticed that "bor", "chlor", "brom", "iod", "fluor" and "phosphor" are used instead of "boron", "chlorine", "bromine", "iodine", "fluorine" and "phosphorus".
>>>
>>>- In the .bat files, there are extra semicolons at the end of the "set CLASSPSTH" instructions.
>>
>>Thanks, i will fix it.
>>
>>
>>
>>>Thanks again.
>>>
>>>Ambrogio
>>
>>
>>Regards, Joerg
>>
>>P.S.: BTW there exists also a descriptor normalization under
>>joelib.process.types.DescVarianceNorm if you plan to normalize your data 
>>with mean=0, sdtdev=1, which requires a little bit of programming. E.g. 
>>you can add this process simply to the selection process pipe in
>>joelib.test.DescriptorSelection
>>
>>P.P.S.: All descriptors in
>>joelib\data\plain\desc2ignore.txt
>>defined in the joelib.properties file as:
>>joelib.process.types.DescVarianceNorm.descriptors2ignore=joelib/data/plain/desc2ignore.txt
>>will be ignored for normalization. This is extremely usefull if you have 
>>  an nominal classification problem or the sdf-file contains also some 
>>id's, ...
>>
> 
> 


-- 
Dipl. Chem. Joerg K. Wegner
Univ. Tuebingen, Computer Architecture, Sand 1, D-72076 Tuebingen, Germany
Tel. (+49/0) 7071 29 78970, Fax (+49/0) 7071 29 5091
E-Mail: mailto:we...@in...
WWW:    http://www-ra.informatik.uni-tuebingen.de