[OTDev] Normalization and PMML
chung chvng at mail.ntua.grThu Nov 5 13:36:37 CET 2009
- Previous message: [OTDev] IBMC services
- Next message: [OTDev] Development Digest, Vol 10, Issue 5
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Dear All, I want to provide PMML representations of SVM models but prior to training I scale the training data so that each attribute has values within [-1,1]. This is done via weka using the class weka.filters.unsupervised.attribute.Normalize. I get the coefficients of the trained model via libSVM and finally I want to create a PMML representation for my model (version 3.2. or higher). Normalization is in fact a linear transformation of the initial data according to a relation: X_k(i) = a_k*x_k(i)+ b_k where X_k is the k-th attribute of the dataset and X_k(i) its i-th element. So, I guess I have to use the element <LinearNorm> of PMML but I cannot understand (and the documentation doesn't help!) if the "norm" stands for the coefficient a_k and if "orig" is the bias b_k. Another thing... Is it correct that the <VectorInstance> elements are the scaled data? Does anybody have any examples for SVM models with scaled data??? Best Regards, Pantelis
- Previous message: [OTDev] IBMC services
- Next message: [OTDev] Development Digest, Vol 10, Issue 5
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Development mailing list