Visualization of the separation plane for the two-variable SVM case

Question

Hello,
I have an training data set for binary classification task that contains only two variables.
After building SVM model (with linear kernel) I want to visualize separation plane.
After reading documentation I found that I can get variable weights using this query:
WITH
mod_dtls AS (
SELECT *
FROM TABLE(DBMS_DATA_MINING.GET_MODEL_DETAILS_SVM('<MODEL_NAME>'))
model_details AS (
SELECT D.class, A.attribute_name, A.attribute_value, A.coefficient
FROM mod_dtls D,
TABLE(D.attribute_set) A
ORDER BY D.class, ABS(A.coefficient) DESC
SELECT class, attribute_name aname, attribute_value aval, coefficient coeff
FROM model_details
WHERE ROWNUM < 100
So, after getting these weight does anybody knows how to plot the separation plane?
How to get equation of the line that separates my two classes?

Answer 1

WITH
mod_dtls AS (
SELECT *
FROM TABLE(DBMS_DATA_MINING.GET_MODEL_DETAILS_SVM('SAMPLE_DATA65805_SV'))
model_details AS (
SELECT D.class, A.attribute_name, A.attribute_value, A.coefficient
FROM mod_dtls D,
TABLE(D.attribute_set) A
ORDER BY D.class, ABS(A.coefficient) DESC
SELECT class, attribute_name aname, attribute_value aval, coefficient coeff
FROM model_details
WHERE ROWNUM < 100 ;
CLASS     ANAME     AVAL     COEFF
0     X          -1.745703462
0               -0.928055641
0     Y          0.813909145
For the linear discriminant the plane separating two classes is -1.745703462x+0.813909145y-0.928055641=0. Or the linear discriminant is y~=2x+1. Unfortunately ODM GUI doesn't provide a method to plot this. In here an important thing to keep in mind that the discriminant function is generated with respect to normalized data (if you choose to do normalization). If you wish to draw some plots by using a self generated method, you should draw discriminant function over normalized data (or you can transform discriminant function).
Regards,
HS