gbev.fit {gbev}R Documentation

Helper function for gbev-function.

Description

Help function for fitting boosted regression trees with errors-in-variables.

Usage


gbev.fit(w,y,
        weights=NULL,
        measErrorModel=NULL,
        method="L2", 
        indepFitUpdate=1,
        nboost=1000,
        lambda=100,
        maxDepth=2,
        m=1,
        minSplit=10, 
        minBucket=0,
        sPoints=10,
        mc=2,
        intermPred=10,
        maxSplitAttempts=10)

Arguments

w Matrix of covariates.
y A vector of responses.
weights Weights applied to observations. Defaults to 1.
measErrorModel This is a list specifying the distribution of the latent covariates and the measurement error. Here it is assumed that the latent covariates are a mixture of normals (possibly multivariate), and that the measurement error is normally distributed. See examples below for details.
method Can be L2 for squared error loss, logLike for binary regression with negative log-likelihood loss.
indepFitUpdate If indepFitUpdate=1 then model fit is updated using independent MC-sample, else the fit is updated using the same MC-sample as used in tree fitting.
nboost Number of boosting iterations to perform.
lambda Regularization parameter.
maxDepth Determines maximum interaction depth of trees fit. maxDepth=2 fits stumps.
m Number of randomly sampled covariates used to split node, usually set to number of covariates.
minSplit Minimum expected number of observations in a node for it to be split.
minBucket Minimum expected number of observations in a node.
sPoints Number of points sampled at random from which to choose split.
mc Number of Monte-Carlo samples drawn to compute node probabilities.
intermPred Increments of iterations at which intermediate predictions are saved, required for cvLoss function.
maxSplitAttempts Maximum number of attempts at finding a valid split point. When splitting a node, sPoints candidate splits are supposed to be found for each covariate, however, each randomly sampled split point does not necessarily give a valid split point (i.e. a point satisfying minBucket and minSplit), and maxSplitAttempts is the maximum number of attempts at finding such a point. .

Details

This function is called by gbev but is not intended for end-user use.


[Package gbev version 0.1.1 Index]