NIR: Near-Isotonic Regression In neariso: Near-Isotonic Regression

Description

These functions are the main interface functions for calculating Nearly-isotonic solutions

Usage

 ```1 2 3``` ```neariso(y, maxBreaks=100, lambda=NULL) nearisoGetBreakpoints(nearisoPathObj, maxBreaks=100) nearisoGetSolution(nearisoPathObj, lambda=nearisoGetBreakpoints(nearisoPathObj)) ```

Arguments

 `y` response variable; numeric `lambda` penalty parameter vector (non-negative) for the difference of coefficients; numeric `nearisoPathObj` Solution object of class `nearisoPath` as returned by nir `maxBreaks` maximum number of breakpoints the function should return

Details

`neariso` is the main function to calculate a Nearly-isotonic regression fit and returns an object of class `nearisoPath`. If lambda=NULL, then the breakpoints of the linear solution path of beta are chosen, however at most `maxBreaks`. See `nirGetBreakpoints` for what happens if there are more than `maxBreaks` breakpoints.

`nearisoGetSolution` takes an object of class `nearisoPath` and returns an object of the same class, but with the solution calculated for the given value of `lambda`. Advantage in comparison to `neariso` is that it uses the already calculated solution and does not recompute the entire solution path, therefore being faster.

`nearisoGetBreakpoints` returns the lambda values at which the piecewise linear solution paths for beta have a breakpoint. If there are more than `maxBreaks` such breakpoints, only `maxBreaks` representative breakpoints will be returned, including the first and last.

Value

Returns a list with the items

 `solObj` Object of class `nearisoSolObj` in which the whole solution path is saved in compact form; used as basis to recalculate fits with new values of `lambda` in `nearisoGetSolution` `lambda` Values of `lambda` for which the solution was calculated `df` Number of different values for beta in the solution; degrees of freedom; same length as `lambda` `beta` `length(y)` x `length(lambda)` matrix with the solution

Holger Hoefling

References

Ryan Tibshirani, Holger Hoefling, Rob Tibshirani. Nearly-isotonic regression. 2010. To appear in Technometrics.

Examples

 ``` 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18``` ```library(neariso) # generate some artificial data y <- rnorm(1000) + (1:1000)/3000 ### run the algorithm as default; will output solution at 100 breakpoints for lambda res0 <- neariso(y) ### apply function nir and get solution directly lambda = 0:10/10 res <- neariso(y, lambda=lambda) ### apply the function and get the solution later res2 <- neariso(y, lambda=NULL) res2 <- nearisoGetSolution(res2, lambda=lambda) ### look at the breakpoints lambdaBreaks <- nearisoGetBreakpoints(res2, maxBreaks=1000) res3 <- nearisoGetSolution(res2, lambda=lambdaBreaks) ```

neariso documentation built on May 1, 2019, 8:21 p.m.