The Altair Community is migrating to a new platform to provide a better experience for you. In preparation for the migration, the Altair Community is on read-only mode from October 28 - November 6, 2024. Technical support via cases will continue to work as is. For any urgent requests from Students/Faculty members, please submit the form linked here
Nearest Neighbor enhancements
Maven
Hi,
I would like to see NearestNeighbor enhanced in the following ways:
1) When computing the nearest neighbor for a given point, optionally exclude points where distance = 0 (most often, this will be just itself). As it is now, if I build a KNN model on a dataset with K=1, and then apply the model to the dataset, the predictions are perfect since the nearest neighbor of each point is itself. This is similar to what Weka's LinearNNSearch option -S does (although the other nearest neighbor algorithms Weka supports unfortunately don't have this option).
2) Be able to specify the weighting kernel function, rather than just have a toggle for"weighted_vote". This would bring it up to the same capabilities as W-LWL, in which can specify linear, Epanechnikov, tricube, inverse, or gaussian weights.
3) Ability to build a full local polynomial regression (aka loess) model, similar to what locfit does in R.
I would like to see NearestNeighbor enhanced in the following ways:
1) When computing the nearest neighbor for a given point, optionally exclude points where distance = 0 (most often, this will be just itself). As it is now, if I build a KNN model on a dataset with K=1, and then apply the model to the dataset, the predictions are perfect since the nearest neighbor of each point is itself. This is similar to what Weka's LinearNNSearch option -S does (although the other nearest neighbor algorithms Weka supports unfortunately don't have this option).
2) Be able to specify the weighting kernel function, rather than just have a toggle for"weighted_vote". This would bring it up to the same capabilities as W-LWL, in which can specify linear, Epanechnikov, tricube, inverse, or gaussian weights.
3) Ability to build a full local polynomial regression (aka loess) model, similar to what locfit does in R.
0
Answers
thanks for sending those suggestions in. Some are easier to implement, others will of course need more time. However, I have added all points to our Todo list.
Cheers,
Ingo