In order to make sure if it is corect , all the instances included in the
model are support vectors right?
so what is the difference between -0.35543718075692615 * k[1282] and -1.0
* k[1283]. what kind of information does the coefficient give?
does weka give the exact number of support vectors in any place?
thank you
çağla
2012/12/21 Çağla Odabaşı <caglajf(a)hotmail.com>
I got it, thank you very much Eibe ....
Subject: Re: [Wekalist] Re: support vectors
From: eibe(a)waikato.ac.nz
Date: Fri, 21 Dec 2012 11:06:54 +1300
To: wekalist(a)list.scms.waikato.ac.nz
Sorry, this should obviously be: support vectors are those instances for which the
coefficient is *NOT* 0.
Note that only instances inside the "tube" whose width is defined by the
epsilon parameter in the epsilon-insensitive loss function will get a coefficient of zero.
This tube is very narrow by default because epsilon is set to 1.0e-3 by default. That
means almost all instances will be support vectors in most cases with the default setting.
To get a sparse solution, you need to increase epsilon, i.e. use something like
java weka.classifiers.functions.SMOreg -t ~/datasets/numeric/fishcatch.arff -K
"weka.classifiers.functions.supportVector.PolyKernel -E 2" -I
"weka.classifiers.functions.supportVector.RegSMOImproved -L 0.1"
In this case epsilon is set to 0.1, giving the tube a width of 0.2.
Cheers,
Eibe
On 21 Dec 2012, at 08:39, Eibe Frank wrote:
Support vectors are those instances for which the
coefficient is 0. They are not included in the output. It looks like (almost) all the
instances in your data are used as support vectors given the parameters you have set for
the SVM.
Cheers,
Eibe
On 20 Dec 2012, at 22:22, Çağla Odabaşı wrote:
Hello
Thanks for your answer, but the problem is I couldnt see the number of support vectors. I
give the end of the result window above and there is no number output for support
vectors. Am I missing ? Or I have 1336 support vectors. It is impossible because my
dataset has 1337 instances. So support vectors may be the instances that have different
coefficients rather then -1 and +1 such as -0.4957011878880593 * k[1286] ?
Thank you
Çağla
-0.35543718075692615 * k[1282]
-1.0 * k[1283]
+0.71130644527556 * k[1284]
+1.0 * k[1285]
-0.4957011878880593 * k[1286]
+1.0 * k[1287]
-1.0 * k[1288]
+1.0 * k[1289]
-1.0 * k[1290]
+0.1830970520846928 * k[1291]
+0.5206676131938187 * k[1292]
-1.0 * k[1293]
-1.0 * k[1294]
+1.0 * k[1295]
+1.0 * k[1296]
+1.0 * k[1297]
+0.902239634793361 * k[1298]
-1.0 * k[1299]
+0.5629567637229693 * k[1300]
+0.3457533296301735 * k[1301]
-0.15914482298907273 * k[1302]
-1.0 * k[1303]
+1.0 * k[1304]
-0.2171786055396634 * k[1305]
-0.042216095699512506 * k[1306]
+1.0 * k[1307]
-1.0 * k[1308]
-0.02239320061684015 * k[1309]
-0.5554609806177595 * k[1310]
+0.5650387929800857 * k[1311]
-0.9802724737246912 * k[1312]
+1.0 * k[1313]
-0.03324819576000365 * k[1314]
-1.0 * k[1315]
-0.26152633089809696 * k[1316]
+1.0 * k[1317]
-0.25371772296571343 * k[1318]
-0.9057193883026416 * k[1319]
+1.0 * k[1320]
+0.9630628285978649 * k[1321]
-1.0 * k[1322]
-1.0 * k[1323]
-1.0 * k[1324]
-0.7686155628772334 * k[1325]
+1.0 * k[1326]
-0.5050959047355967 * k[1327]
-0.7736451077957822 * k[1328]
+1.0 * k[1329]
-1.0 * k[1330]
+1.0 * k[1331]
-0.6939227835732249 * k[1332]
-1.0 * k[1333]
+1.0 * k[1334]
+1.0 * k[1335]
-0.9135711091795383 * k[1336]
+ 0.4497
Number of kernel evaluations: 2704413 (94.438% cached)
Time taken to build model: 2.44 seconds
=== Cross-validation ===
=== Summary ===
Correlation coefficient 0.9134
Mean absolute error 9.4922
Root mean squared error 15.019
Relative absolute error 28.8441 %
Root relative squared error 40.9704 %
Total Number of Instances 1337
From: mhall(a)pentaho.com
To: wekalist(a)list.scms.waikato.ac.nz
Date: Thu, 20 Dec 2012 01:49:00 -0600
Subject: Re: [Wekalist] Re: support vectors
From: çağla odabaşı <caglajf(a)gmail.com>
Reply-To: "Weka machine learning workbench list."
<wekalist(a)list.scms.waikato.ac.nz>
Date: Wednesday, 19 December 2012 5:25 AM
To: "Weka machine learning workbench list."
<wekalist(a)list.scms.waikato.ac.nz>
Subject: [Wekalist] Re: support vectors
2012/12/18 çağla odabaşı <caglajf(a)gmail.com>
Hello
I am using SMOreg in weka and I have to know how many support vectors were used to make
the model. Does weka give information about it? I can see the model and count the support
vectors from there, but the model is too long.
The vectors are numbered in the output, so scrolling to the end of the model will tell
you how many were used.
Cheers,
Mark.
_______________________________________________ Wekalist mailing list Send posts to:
Wekalist(a)list.scms.waikato.ac.nz List info and subscription status:
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist List
etiquette:http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist(a)list.scms.waikato.ac.nz
List info and subscription status:
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette:
http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
_______________________________________________ Wekalist mailing list Send
posts to: Wekalist(a)list.scms.waikato.ac.nz List info and subscription
status:
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist List
etiquette:
http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
_______________________________________________
Wekalist mailing list
Send posts to: Wekalist(a)list.scms.waikato.ac.nz
List info and subscription status:
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette:
http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html