osurdml / issues

0 stars 0 forks source link

Estimate Treemower score with Gaussian Process Regression #25

Open yoos opened 10 years ago

yoos commented 10 years ago

Currently, the score is simply a sum of the scores of all traversed cells. We should instead implement a Ordinary Kriging evaluator in DecisionTree::CalcScore().

drews256 commented 10 years ago

We could process the data in Matlab. As long as the data output is something that I can use or I can make it into something that I can use then I should be able to do that. I'll play around with it tonight.

yoos commented 10 years ago

Sounds good. We currently export the generated path to a logfile in the output directory. Between the original costmap and the score depreciator in DecisionTree::DepreciateScore(), I think you'll have the information you need.

drews256 commented 10 years ago

Also there is a difference between "simple" kriging and ordinary kriging. I don't know if you meant "simple" kriging as in a simple evaluator that uses kriging or a "simple kriging" in the sense that we're using the simple kriging technique. I think ordinary kriging is more applicable.

yoos commented 10 years ago

Oh, I saw a mention of Simple Kringing in your paper, thus my assumption. Ordinary kringing it is.

drews256 commented 10 years ago

It sounds like you're on board with this anyways so I will be working towards that goal.

drews256 commented 10 years ago

"Even so, they are useful in different frameworks: Kriging is made for estimation of a single realization of a random field, while regression models are based on multiple observations of a multivariate data set." -Per Wikipedia. Maybe we should just call it Gaussian Process Regression

yoos commented 10 years ago

We're going to be mathing on multiple multivariable mathing maths.

drews256 commented 10 years ago

Yes, m'ing on quadruple m's is really the way to do it.

yoos commented 10 years ago

Mmm-mm. Can't wait.