Result from Write Model different from Kernel Model?

Fred12

hi,

I want to store my SVM Model, when I do "store Kernel Model" it seems to be a different result in the model than when I do write Model. In the last case, I get a big ".mod" file, whereas in s tore Kernel Model, I get an IOO or ".md" object...?

for the correct model, do I always have to do "write Model" ?? 

bhupendra_patil

hi Fred, these are just various formats, under the hood its the same model,

 

Store lets you save it in a repository

 

write lets you write outside of a repo location

 

Use what suits your needs, I'll recommend using Store wherever possible

Fred12

ok but I am getting totally different results...

I got once 12% from my SVM Model, but it somehow had not the right format either:

I had 3 classes trained before : 1.0, 5.0 and 10.0,  however, I get a contingency table with 6 entries:  1 , 5, 10 and 1.0 , 5.0 and 10.0 , I will post a screenshot later when I'm at home...

is there somehow an issue with integer or real data types?

 

with the *.mod model it gives me right contingency table with correct classes, but worse performance (70% compared to 85%) on the test-set in the process from x-validation...

 

edit:

this is how my process looks for testing the model:

 

 

 

I trained my model on nominal data, when I test it on same dataset, but this time with polynomial data type format (for the label), I get this result:

 I guess that's a bug ?

 

 

when I test it with nominal data type dataset, I get this result:

 

 

but 70% are still far worse than my 86% that I achieved in X-validation.. why??