Cool.. I'm looking at it now. And I remembered that the other option was to leave it in and store the weights in the model. This would slim down the classification process a bit. And keep complementary naive Bayes from calling computeWeight() twice.. Once during training (in the thetaMapper) and again in testing/classification. I'm not sure there's a reason not to store complete weights during training.. will look into it.
Sent from my Verizon Wireless 4G LTE smartphone -------- Original message -------- From: Sebastian Schelter <[email protected]> Date:04/20/2014 11:57 AM (GMT-05:00) To: [email protected] Subject: Re: [jira] [Created] (MAHOUT-1519) Remove StandardThetaTrainer Happy to see you take it! On 04/20/2014 05:52 PM, ap.dev wrote: > I can do this later today if u want if u haven't started on it ... assuming I > get Internet back. > > > Sent from my Verizon Wireless 4G LTE smartphone > > -------- Original message -------- > From: "ap.dev" <[email protected]> > Date:04/20/2014 11:37 AM (GMT-05:00) > To: [email protected] > Subject: RE: [jira] [Created] (MAHOUT-1519) Remove StandardThetaTrainer > > Responded to wrong email before... > > @ssc, yes theta normalization job is only needed for CBayes. > StandardThetaTrainer can come out. Running the thetaNormalization job for > Bayes is an Unneeded expense. > > There is a preconditions check in Bayes model that looks for non zero values > in the perlabelThetaNormalizer vector in model.validate () but other than > that it car be I don't think that it is referenced at all. > > My Internet is down this morning so i can't get into jira.. will log in > whenever it comes back up. > > > Sent from my Verizon Wireless 4G LTE smartphone > > > -------- Original message -------- > From: "Sebastian Schelter (JIRA)" > Date:04/20/2014 7:23 AM (GMT-05:00) > To: [email protected] > Subject: [jira] Sebastian Schelter mentioned you (JIRA) > > > [ > https://issues.apache.org/jira/browse/MAHOUT-1519?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel > ] > > Sebastian Schelter mentioned you on MAHOUT-1519 > --------------------------------------- > > [~Andrew_Palumbo] if I understand your work in MAHOUT-1504 correctly, the > theta training is only necessary for complementary naive bayes, right? > Then, we should remove the StandardthetaTrainer and make the > TrainNaiveBayesJob only do the theta training in the complementary case. > > Correct me if I miss something here. > >> Key: MAHOUT-1519 > >> View Online: https://issues.apache.org/jira/browse/MAHOUT-1519 >> Add Comment: >> https://issues.apache.org/jira/browse/MAHOUT-1519#add-comment > > Hint: You can mention someone in an issue description or comment by typing > "@" in front of their username. > > > > > -- > This message was sent by Atlassian JIRA > (v6.2#6252) > > > Sent from my Verizon Wireless 4G LTE smartphone > > -------- Original message -------- > From: "Sebastian Schelter (JIRA)" <[email protected]> > Date:04/20/2014 7:23 AM (GMT-05:00) > To: [email protected] > Subject: [jira] [Created] (MAHOUT-1519) Remove StandardThetaTrainer > > Sebastian Schelter created MAHOUT-1519: > ------------------------------------------ > > Summary: Remove StandardThetaTrainer > Key: MAHOUT-1519 > URL: https://issues.apache.org/jira/browse/MAHOUT-1519 > Project: Mahout > Issue Type: Improvement > Components: Classification > Reporter: Sebastian Schelter > Fix For: 1.0 > > > [~Andrew_Palumbo] if I understand your work in MAHOUT-1504 correctly, the > theta training is only necessary for complementary naive bayes, right? > Then, we should remove the StandardthetaTrainer and make the > TrainNaiveBayesJob only do the theta training in the complementary case. > > Correct me if I miss something here. > > > > -- > This message was sent by Atlassian JIRA > (v6.2#6252) >
