libagf-general — General--all topics

[Libagf-general] New crowd-sourced project

[Peter M. <pet...@ho...>]

Dear libAGF'ers:

Please support my new crowd-sourced research project:
http://www.iamscientist.com/projects/12-Isoline-retrieval-of-ozone-in-the-stratosphere

Thanks,

Peter

Peter Mills
Maverick scientist and software developer

http://sourceforge.net/users/peteysoft/
-Innovative, original and efficient numerical algorithms

http://libagf.sourceforge.net/
-Fast, simple and accurate statistical classification.
 		 	   		  

[Libagf-general] New release of Adaptive Gaussian Filtering

[Peter M. <pet...@ho...>]

libagf version 0.9.4 is now available on Sourceforge.  The major change in this version is that all subroutines, except for the I/O, have been templated.  This, combined with typedef'ing all variables in the main routines allow you to easily switch between single-precision and double-precision for the floating point operations and between different size integers for the class labels.

Also new is an experimental pdf estimation method with optimal error rates.  The final, operational version, should be ready in the next release.

And coming soon as well: multi-class classification using the class-border method.  Stay tuned!

For all news related to libagf, please go to the sourceforge page:
https://sourceforge.net/projects/libagf/
or the libagf webpage:
http://libagf.sourceforge.net

You might also want to check out some of our other projects at:
http://peteysoft.org

Happy classifying,

Peter

Peter Mills
Maverick scientist and software developer

http://sourceforge.net/users/peteysoft/
-Innovative, original and efficient numerical algorithms

http://libagf.sourceforge.net/
-Fast, simple and accurate statistical classification.
 		 	   		  

[Libagf-general] Patches for libagf-0.9.2‏

[Peter M. <pet...@ho...>]

Dear Ernst:
 
It would be nice if you could submit your patches via cvs, but I realize not everyone has the time, inclination or (in your case) resources to become a developer.  Thanks a lot for the heads up on the bugs.  I was aware of the problem with longword integers on 64 bit machines--at the time I started programming I didn't realize there was a difference.  
 
As long as you create your input files on the same machine as you run the codes, there usually isn't a problem, but as you point out, at minimum it should be mentioned in the documentation.  I was considering making the file formats more portable, but I appreciate the speed and simplicity of straight binary files.  This way, the main routines can be easily patched in to other programming languages as simple system calls with very little speed penalty.  
 
For maximum flexibility, types could be specified as template declarations while in the main routines, integers could be defined consistently across different machine architectures.  I hope the next version of the codes will have this feature.
 
Could you please submit your patches to the mailing list I just created?  Thanks in advance!
And thanks all for contributing to libagf!
 
Kind regards,

Peter

Peter Mills Maverick scientist and software developer http://sourceforge.net/users/peteysoft/ -Innovative, original and efficient numerical algorithms http://libagf.sourceforge.net/ -Fast, simple and accurate statistical classification. 


 
> To: pet...@us...
> From: er...@us...
> CC: 
> Subject: possible (small) contribution to libagf
> Date: Tue, 14 Dec 2010 22:08:11 +0000
> 
> Hello,
> 
> while trying to compile libagf (0.9.2) with g++4.4 on a
> linux system (32bit) there were some difficulties up to
> program crashes. I made some small changes / fixes in the
> code that solve this.
> 
> Also I tried to compile on a 64bit machine, which also
> required some fixes. The code now compiles and runs,
> including the tests. But the 64bit version is not yet
> correct as the content of the binary data files written is
> different from the documentation (long integers are 8byte
> with this compiler on 64bit). Still the code changes I made
> might be of some use in the future.
> 
> Are you interested in receiving patches? I could create
> patches against the content of libagf0.92.tgz and send them
> by mail. I do not have cvs access from work and can not do
> anything from home. 
> 
> Ernst
> 
> --
> This message has been sent to you, a registered SourceForge.net user,
> by another site user, through the SourceForge.net site. This message
> has been delivered to your SourceForge.net mail alias. You may reply
> to this message using the "Reply" feature of your email client, or
> using the messaging facility of SourceForge.net at:
> https://sourceforge.net/sendmessage.php?touser=3164728
> 
 		 	   		  

[Libagf-general] Patches for libagf-0.9.2

[Peter M. <pet...@ho...>]

Dear Ernst:
 
It would be nice if you could submit your patches via cvs, but I realize not everyone has the time, inclination or (in your case) resources to become a developer.  Thanks a lot for the heads up on the bugs.  I was aware of the problem with longword integers on 64 bit machines--at the time I started programming I didn't realize there was a difference.  
 
As long as you create your input files on the same machine as you run the codes, there usually isn't a problem, but as you point out, at minimum it should be mentioned in the documentation.  I was considering making the file formats more portable, but I appreciate the speed and simplicity of straight binary files.  This way, the main routines can be easily patched in to other programming languages as simple system calls with very little speed penalty.  
 
For maximum flexibility, types should be specified as template declarations while in the main routines, integers should be defined consistently across different machine architectures.  I hope the next version of the codes will have this feature.
 
Could you please submit your patches to the mailing list I just created?  Thanks in advance!
And thanks all for contributing to libagf!
 
Kind regards,
 
Peter
 
Hello,
 
while trying to compile libagf (0.9.2) with g++4.4 on a
linux system (32bit) there were some difficulties up to
program crashes. I made some small changes / fixes in the
code that solve this.
 
Also I tried to compile on a 64bit machine, which also
required some fixes. The code now compiles and runs,
including the tests. But the 64bit version is not yet
correct as the content of the binary data files written is
different from the documentation (long integers are 8byte
with this compiler on 64bit). Still the code changes I made
might be of some use in the future.
 
Are you interested in receiving patches? I could create
patches against the content of libagf0.92.tgz and send them
by mail. I do not have cvs access from work and can not do
anything from home. 
 
Ernst
 


Peter Mills Maverick scientist and software developer http://sourceforge.net/users/peteysoft/ -Innovative, original and efficient numerical algorithms http://libagf.sourceforge.net/ -Fast, simple and accurate statistical classification.