Re: [Jboost-users] JBoost questions

[David R. <rol...@ya...>]

Hi Aaron,=0A=0AI'm starting a new thread because I never received your repl=
y through my mail, so I can't reply to the thread.=0A=0AI just downloaded v=
ersion 1.4, I'll take a look at it really soon. I'm still in the process of=
 evaluating JBoost to see if it fits my needs. So any change to the ouput f=
iles format doesn't affect me right now. The GUI will be a very interesting=
 tool, even though I already made my own batch files to call JBoost.=0A=0AY=
our explanation of score/margin is clear, but there is still something that=
 I don't understand. The graphic for the margin has "Margin" and "Cumulativ=
e Distribution" as axis labels. In this graphic, the margin values are betw=
een -1 and +1. This means that the values have been remapped from the formu=
la you gave me (margin =3D score * label). Am I right ? This remap was actu=
ally the purpose of my initial question, but I wasn't clear on that point.=
=0A=0AI began using Adaboost for image analysis (mainly OCR) where I was wo=
rking, but seeing it's interesting performances, I want to try it in other =
domains. I'll give you feedback it it works well. From my previous experien=
ce with OCR, Adaboost isn't magical, the input data must be chosen carefull=
y and right now I'm working on the input before really using JBoost.=0A=0A =
Thanks for the previous answers,=0A=0ADavid R.=0A=0A=0A--------------------=
----=0A=0AHi David,=0A=0ASee responses. Also know that for release 1.4 (com=
ing out sometime this  =0Aweek most likely) we've completely changed the fo=
rmat of the output files  =0Aand post-processing. We are also in the proces=
s of porting all  =0APython/Perl code to Java. We are also developing a GUI=
 to make the whole  =0Aprocess more intuitive, which will probably be relea=
sed in the next month.  =0AThat being said...=0A=0AOn Mon, 15 Oct 2007, Dav=
id Rolland wrote:=0A=0A> First, I tested the visualization tools to make su=
re I had the same  =0A> results as shown on  =0A> http://www.cs.ucsd.edu/~a=
arvey/jboost/doc.html#visualization. I had to  =0A> modify the atree2dot2ps=
.pl script because it does not use the "--dir"  =0A> option everywhere in t=
he script. Is it the expected behavior ?=0A=0AYou're right. The $dirname va=
riable is only used for the $infofilename,  =0Anot the $filename. I've comm=
itted the change to CVS. See  =0Ahttp://sourceforge.net/cvs/?group_id=3D195=
659 for details on how to get CVS  =0Aaccess.=0A=0A> Second, I was not able=
 to reproduce the margin output. My problem is  =0A> that I run JBoost on W=
indows XP and the margin script uses 'cat' and  =0A> other Unix commands.=
=0A=0AYeah... the scripts were written assuming a few other things too...  =
=0AReally, they were written as bandages, not final solutions. We're  =0Acu=
rrently working on porting all of this to Java so that we have more  =0Aint=
eroperability.=0A=0A> I am a C/C++ programmer, I don't know Perl nor Python=
.=0A=0APerl is pretty out of style these, though it is amazing what some of=
 that  =0Agross syntax can do. Python has survived a couple of fads (ruby, =
etc) and  =0Astill seems to me the best widespread scripting language. It m=
ay be worth  =0Aa look, even if not for this project.=0A=0A> Even though th=
ese languages are similar to C I still wonder how the  =0A> files spambase.=
train.margin and spambase.train.scores can be processed  =0A> to output the=
 margin graphic.=0A=0AWe use gnuplot as an intermediary...=0A=0A> And actua=
lly, what's the difference between these two files?=0A=0AOne has the "score=
" of the example, the other has the "margin". The score  =0Ais defined as t=
he value predicted for a given example. The margin is the=0A=0Aif (label of=
 example is correct)=0Areturn |score|=0Aelse=0Areturn - |score|=0A=0Awhere =
|x| is the absolute value of x.=0A=0A> There only seems to be some positive=
/negative changes of values. I don't  =0A> expect you to rewrite margin.py =
without taking advantage of Unix  =0A> commands, but if you explain how I c=
an get from the data files to the  =0A> output graphic, I'll code myself a =
Windows equivalent.=0A=0AI'd say the best bet for the moment being is to gr=
ab cygwin, where  =0Aeverything has been tested and *seemed* to work peachy=
. The second  =0Abest bet would probably be to wait till Wednesday/Thursday=
 for the next  =0Arelease (when all of your changes would probably be rende=
red somewhat moot  =0Aanyways). Third best bet is edit the code.=0A=0AThe o=
nly place where UNIX commands are used in 1.3.1 margin.py are lines  =0A244=
--255. That is where a label file is created. All labels in JBoost  =0Aare =
converted into binary values +1, -1. If you have more than two  =0Alabels, =
just wait till later this week. if you have two labels, then just  =0Afigur=
e out which is mapped to "+1" and which to "-1" (should be fairly  =0Astrai=
ght forward). Create the labels file, which is just a series of 1,  =0A-1 r=
ead in. The formula for margin is (line 168)=0A=0Amargin =3D score * label=
=0A=0AThis is identical to what I state above for when +1,-1 are used. So i=
f  =0Ayou write a script to create a label file, you can specify the file o=
n the  =0Acommand line (via --labels=3D...), and all your problems should b=
e solved.=0A=0ALet me know if you have any other questions/comments.=0A=0AA=
lso, out of curiosity, for what classification task are you using JBoost?=
=0A=0AAaron=0A=0A=0A=0A=0A      Be smarter than spam. See how smart SpamGua=
rd is at giving junk email the boot with the All-new Yahoo! Mail at http://=
mrd.mail.yahoo.com/try_beta?.intl=3Dca=0A