Regression with Junto

[afshin...@gmail.com]

Hello,

I have a regression problem and the samples are connected in a graph. I use Junto classification for this regression problem by discretizing the real-valued output space and use these discrete intervals as classes. (instead of real values from 1 to 10 i discretize them into intervals of 1-2, ..., 9-10 and then use them as class labels)

Is there anyway i can modify Junto to optimize a regression problem so that i don't have to discretize the output space?

If not, do you know any starting point (papers/implementations) for graph-based semi-supervised regression?

Thank you for your great software.

[Partha Pratim Talukdar]

On Thu, Dec 18, 2014 at 2:36 PM, <afshin...@gmail.com> wrote:

I have a regression problem and the samples are connected in a graph. I use Junto classification for this regression problem by discretizing the real-valued output space and use these discrete intervals as classes. (instead of real values from 1 to 10 i discretize them into intervals of 1-2, ..., 9-10 and then use them as class labels)

Is there anyway i can modify Junto to optimize a regression problem so that i don't have to discretize the output space?

Actually, there is nothing in Junto which prevents you from using it in regression problems. So, you might just want to use real-valued output as label scores and run Junto as is. The final label scores are again real-valued.

Partha