Re: ARFF vs CSV
Eibe Frank <[email protected]>
| Newsgroups | gmane.comp.ai.weka |
|---|---|
| Message-ID | <CADehzLUyj1GvadJcVkdvwVn+oth+cbUK9X7J1nBRyWv7=jeFgA@mail.gmail.com> |
As Peter has mentioned, by default, the ARFF writer in WEKA rounds numeric values to six decimal places. When saving an ARFF file using the standard dialog, there should be a tickbox that allows you to bring up a window with additional options to configure when writing the file (including numeric precision). Similarly, you can manually specify the order of nominal values when reading a CSV file so that it is consistent with the ARFF file, by invoking the dialog with optional settings for the CSV reader. If you can give more details on the algorithms and evaluation methods you used, we may be able to give more insight into why there are differences. In most cases, the differences should not be statistically significant. However, because WEKA generally does not randomly break ties, the order of nominal values may sometimes have a small but statistically significant effect depending on the algorithm used. Of course, if tiny differences of numeric values are important for discrimination, then the lack of numeric precision can have a substantial (and statistically significant) effect. Cheers, Eibe On Sat, 13 Nov 2021 at 14:08, Szucs, Roman <[email protected]> wrote: > I am getting different results depending on whether I open the training > set as .arff file or .csv file. > > > > > > Is this normal? Many thanks > > > > Roman > _______________________________________________ > Wekalist mailing list -- [email protected] > Send posts to [email protected] > To unsubscribe send an email to [email protected] > To subscribe, unsubscribe, etc., visit > https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz > List etiquette: > http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html > _______________________________________________ Wekalist mailing list -- [email protected] Send posts to [email protected] To unsubscribe send an email to [email protected] To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html