search

pierreroudier / clhs

A R implementation of the conditioned Latin Hypercube Sampling method
12 stars 9 forks source link

Implement the DLHS variant method (Minasny and McBratney, 2010) #9

Closed BenjaminLouis closed 6 years ago

BenjaminLouis commented 6 years ago

Hello Pierre,

Here is a PR that adds the DLHS variant method proposed by Minasny and McBratney (2010, see attached). This method is the same that the cLHS but gives more importance to the edge of the distributions, inspired by the D-optimality in linear regression.

The changes are not huge as it just implement a new parameter of the objective function, called eta, which is a matrix of size N x K (where N is the size of the sample et K the number of continuous variable in the data frame). This matrix allows to give more weight for chosen strata of the distributions (i.e. the edge) in the sense that the probability of sampling in these strata is increased. I implemented the parameter as a matrix (instead of one integer as explained in the article) to allow more complex weight design.

I hope this PR will fit the purpose of your package. Thanks for considering it. Feel free to contact me (@BenjaminLouis) if you have questions.

King regards

Minasny&McBratney2010.pdf

pierreroudier commented 6 years ago

@BenjaminLouis Thanks a lot for the PR!

I will add your name to the contributors' list in DESCRIPTION -- if that's ok with you.

pierreroudier commented 6 years ago

@BenjaminLouis Could you please write a simple test of this feature?

BenjaminLouis commented 6 years ago

@pierreroudier I will do it during the weekend