forager_implementation_notes.md

In talegari/forager: Compute auxiliary information (proximity, dissimilarity, outlyingness, depth) and imputation from tree ensembles on new data

Implementation Checklist

package: forager

Legend:

• ??Not sure
• TODO: Idea is clear, to be implemented. Without a name refers to Srikanth.
• DONE: done

Design

• forage should be called with a 'what'
• ex:
  • forage(model, data, what = "dist", ...)
  • forage(model, data, what = "outlier", ...)
• Add plot methods when necessary (low priority)
• Support modelList (low priority)

Whats

• dist (TODO, dependency on parallelDist PR)

  • ranger/randomForest: Use hamming from 'parallelDist' package

• simil (TODO)

  • from dist

• knn (TODO)

• frnn (TODO)

• Outlier detection

• method:

  • proximity:

    • dist: (DONE with dependency on dist / knn / frnn)

    • knn/frnn: (TODO)

      • uncertainity: (mostly TODO Radha)

    • regression:

      • quantreg:

        • ranger: distance from median / length(prediction interval)

        • quantregForest: do the same (see what else it implements) - rf:

        • ranger: Use infjk or jk - randomForest: ??Implement this (TODO)

• prediction intervals: (TODO) closely follows Outlier detection

• margin: (TODO) classification only

• tree correlation: (TODO) mtry, bias-variance (reg only) from wager's paper

plot methods

TODO

talegari/forager documentation built on May 3, 2019, 4:01 p.m.