Re: Why weka.classifiers.functions.LibSVM creates different results on same dataset?
Emerelle <Emerelle24 <at> yahoo.de>
2012-08-21 07:47:37 GMT
I just take a look at the "-model" option and it works with svm-predict.
However I already created a lot of models with the "-d" option. Is it
possible to convert them or do I really have to retrain my models and save
with the new option?
I tried to use following command line
weka.Run LibSVM -l existingModel.model -Test.arff -model
but I get an error. I guess I can't use "-l" "-model" at the same time?
Mark Hall-9 wrote:
> On 15/08/12 12:18 AM, Emerelle wrote:
>> thank you for your fast answer. I just tried it out with the nightly
>> snapshot and works perfectly.
>> Another thing that makes me wonder is why the output model which is
>> by weka can't be used with svm-predict from LibSVM. Is there a was to
>> it compatible?
> The LibSVM package for the developer version of Weka (>= 3.7.2) has a
> "-model" option that allows the internal libsvm model to be saved out to
> a file. This should be compatible with svm-predict.
>> Mark Hall-9 wrote:
>>> On 14/08/12 10:06 PM, Emerelle Queen wrote:
>>>> I was just wondering why I get different results though I use the same
>>>> settings of the LibSVM and the same dataset?
>>>> I am using following command in the commandline:
>>>> weka.classifiers.functions.LibSVM -S 0 -K 1 -D 1 -G 0.0 -R 0.0 -N 0.5
>>>> 40.0 -C 0.5 -E 0.001 -P 0.1 -B -Z -T "testfile" -t "trainfile"
>>>> It would be great if you can explain it to.
>>> Newer versions of the libsvm library (>= 3.1) allow a random seed to be
>>> set, which makes the results deterministic from run to run. LibSVM
>>> (package for Weka >= 3.7.2) and stable-3-6 has been updated with a -seed
>>> option. This is available in the nightly snapshots of Weka 3.6 and will
>>> be in Weka 3.6.8 due out later this week.
>>> Wekalist mailing list
>>> Send posts to: Wekalist <at> list.scms.waikato.ac.nz
>>> List info and subscription status:
>>> List etiquette:
> Wekalist mailing list
> Send posts to: Wekalist <at> list.scms.waikato.ac.nz
> List info and subscription status:
> List etiquette:
View this message in context: http://old.nabble.com/Why-weka.classifiers.functions.LibSVM-creates-different-results-on-same-dataset--tp34295962p34327913.html
Sent from the WEKA mailing list archive at Nabble.com.
Wekalist mailing list
Send posts to: Wekalist <at> 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