1

In my project indoor location I use SVR Regression.

I need to train SVR offline using x,y location and received signal strength WiFi (to build fingerprint database) for training in my location, and build a model after that predict a new location for new signal strength will come online phase.

In SVR Code published predict function in SVR takes an array of numbers as input and returns ONE numeric value as output. But I have two instances, for X and Y.

public double predict(T x);{
}

My question :

Can I train two dimensional x,y together or is each coordinate considered as one output of SVR and trained independently ?

Ashraf Sayed
  • 11
  • 2

1 Answers1

1

You can use both dimension together. Technical term for these dimension would be feature.
Just like one dimension values, you can train your SVR for any number of dimensions. For this your mathematical operation would change accordingly.

Like X +/- Y would be euclidean distance between them .
For multiplication, you could use dot product between them.

If you are using any library of SVR, then most of them task input in 2-d array format having size m*n.

where,
m = number of samples
n = number of dimensions (2 in your case)

saurabh agarwal
  • 2,124
  • 4
  • 24
  • 46
  • Now i use cluster algorithm Affinity Propogation and store orientation (north south east west) as feature then I will train each one from 4 cluster using svr. when I work in online phase I will recieve signal strength and I need predict location. I must calculate euclidean distance between coordinate why? I you have any sample java code help me. I store all strength from all available access point for same location and my environment and it have some noise data for that I must use non linear transformation using kernek function and regression because I have real value data. – Ashraf Sayed Aug 25 '15 at 21:42
  • its very hard to understand what you are asking ? If you are looking for a sample code, you can start with libsvm library. http://stackoverflow.com/questions/10792576/libsvm-java-implementation. – saurabh agarwal Aug 26 '15 at 20:13
  • My answer assumed that you know how to implement svr with 1 dimensional data and you want to know, how to go about two dimensional or n dimensional data. However, its not hard and fast rule to use euclidean distance but its used mostly in practice to compute distance between two points. These points can be of any dimension. – saurabh agarwal Aug 26 '15 at 20:16