You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@mahout.apache.org by "Richard Scharrer (JIRA)" <ji...@apache.org> on 2014/04/26 05:58:15 UTC
[jira] [Updated] (MAHOUT-1525) train/validateAdaptiveLogistic
[ https://issues.apache.org/jira/browse/MAHOUT-1525?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Richard Scharrer updated MAHOUT-1525:
-------------------------------------
Affects Version/s: (was: 0.7)
0.9
> train/validateAdaptiveLogistic
> ------------------------------
>
> Key: MAHOUT-1525
> URL: https://issues.apache.org/jira/browse/MAHOUT-1525
> Project: Mahout
> Issue Type: Question
> Components: Classification
> Affects Versions: 0.9
> Reporter: Richard Scharrer
> Labels: adaptiveLogisticRegression,, newbie
>
> Hi,
> I tried to use train- and validateAdaptiveLogistic on my data which is like:
> category, id, var1, var2, ...var72 (all numeric)
> I used the following settings:
> mahout trainAdaptiveLogistic --input resource/trainingData \
> --output ./model \
> --target category --categories 9 \
> --predictors a0 a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 .....
> --types numeric \
> --passes 100 \
> --showperf \
> mahout validateAdaptiveLogistic --input resource/testData --model model --confusion --defaultCategory none
> The output of validateAdaptiveLogistic is:
> Log-likelihood:Min=-5.54, Max=-0.04, Mean=-1.58, Median=-1.33
> =======================================================
> Confusion Matrix
> -------------------------------------------------------
> a b d e f g h i <--Classified as
> 14 0 0 0 0 0 0 0 | 14 a = projekt
> 0 18 0 0 0 0 0 0 | 18 b = news/aktuelles/presse
> 0 0 24 0 0 0 0 0 | 24 d = lehrveranstaltung
> 0 0 0 19 0 0 0 0 | 19 e = publikation
> 0 0 0 0 20 0 0 0 | 20 f = event
> 0 0 0 0 0 14 0 0 | 14 g = mitarbeiter/person
> 0 0 0 0 0 0 44 0 | 44 h = übersicht
> 0 0 0 0 0 0 0 13 | 13 i = institut
> (in case you were wondering, the categories a in german)
> My problem is that this is impossible. I always get a perfect classification even with just a little amount of training data. It doesnt even matter how many features I use I tried it with all 72 and with only one. Am I missing something?
> Regards,
> Richard
--
This message was sent by Atlassian JIRA
(v6.2#6252)