weighted distance

[naarkhoo]

I enjoy using the package and wonder if you can add the weight feature into the NN2 function where distances are normalised/adjusted according the weights ? (meanwhile, if there is any hack I can do, I appreciate any comments)

[jefferis]

@naarkhoo thanks for your interest int he package. I am not exactly sure what weighting scheme you had in mind – perhaps you can explain. It is not so likely that I can add this since the underlying ANN library has no support for any weighting feature.

[naarkhoo]

In my dataset rows are customers and columns are covariates. I have three covariates, 1)age, 2) category of movie, 3) wether they have bought a movie for someone else. The variance of covariate 3, is much lower, 95% of cases 0 and a few are 1 and the variation in 1 is higher.

So I am thinking of a scheme where one can give weight to each covariate when calculating the distances among rows(customers). If I calculate the distance as it is, I am assuming all covariates are equally important, which I don't think is the case. I believe, this package https://cran.r-project.org/web/packages/distances/distances.pdf provides a weight parameter and has some good examples. Weights either can be based on the variance of each covariance, or manually set.

I hope, I have explained it well - let me know if there is any think unclear. Thank you again.

[jefferis]

@naarkhoo thanks for the explanation. However this normalisation is something that you need to do before you pass your points to RANN::nn2, by scaling each column appropriately. This is exactly what the distances package seems to do. So I am sorry, but it doesn't make sense for RANN to do this.