Hi Adi, The model of the Vertical Hoeffding Tree that is used to predict is inside ModelAggregatorProcessor. It consists basically of the tree structure of nodes, that are learning nodes and splitting nodes. In LocalStatisticsProcessor, we maintain the information needed to decide if we want to do new splits or node. There are two options:
- store only the information inside ModelAggregatorProcessor, so that you can do predictions - store also the information inside LocalStatisticsProcessor, to be able to continue training If you don't need to continue updating the model, then the first option will be fine. Cheers, Albert On Wed, Dec 16, 2015 at 11:45 AM, Albert Bifet <[email protected]> wrote: > I'm redirecting the email to the Samoa developer list. > > On Wed, Dec 16, 2015 at 10:33 AM, Adi Teja Dasari < > [email protected]> wrote: > >> >> >> Hi, >> >> I am Adi, a software engineer and we are using Apache Samoa in one of >> our personal projects. I will get straight to the point… In machine >> learning we have data and we do testing and training which happens in >> *VeticalHoeffdingTree.java >> (follow the link below)* . Once we do training we get a model hypothesis >> and that state might be object of *LocalStatisticsProcessor* , >> *locStatProc* I suppose which will be sent to *ModelAggregatorProcessr* >> through *topologyBuilder.connectInputAllStream(computeStream, >> modelAggrProc);. *We are looking for a way to save that state and >> continue with testing using our data and that state. I will be grateful if >> you could you please help me here if this is the model hypothesis state and >> how do I save it to Hbase. >> >> >> >> >> https://github.com/yahoo/samoa/blob/master/samoa-api/src/main/java/com/yahoo/labs/samoa/learners/classifiers/trees/VerticalHoeffdingTree.java >> >> >> >> Thanking you, >> >> Adi Teja Dasari, >> >> Associate Software Engineer, >> >> Tricon Infotech pvt ltd. >> >> >> ------------------------------ >> Disclaimer: This e-mail is bound by the terms and conditions described at >> http://www.triconinfotech.com/mail-disclaimer/ >>
