Compute the average tree size in a forest

Description

Returns the average number of terminal nodes for trees in a forest

Usage

avSize(forest)

Arguments

Value

The average number of terminal nodes in forest

Examples

avSize(forest)

Flattens predictor variable data into one row per person

Description

Assumes that splitting explanatory variables do not vary with time. Spline Tree is not meant to handle time-varying covariates.

Usage

flatten_predictors(idvar, data)

Arguments

Value

A wide format dataset with spline coefficients as the responses.

Sample forest used in vignettes

Description

Sample forest used in vignettes

Usage

forest

Format

An object of class list of length 15.

Get the basis matrix to be used for this spline tree

Description

Using the user-specified parameters or the default parameters, computes the basis matrix that will be used for building the tree.

Usage

getBasisMat(yvar, tvar, idvar, data, knots = NULL, df, degree, intercept,
  gridPoints, nGrid = 7)

Arguments

Value

The basis matrix to be used for the tree building process

Retrieve the subset of the data found at a given terminal node

Description

Given a terminal node number, this function returns the data belonging to this terminal node. If the dataType argument is 'all', returns all rows of data from the original dataset that fall in this node. Otherwise, the flattened data that belongs to this node is returned (one row of data per ID, original responses replaced by spline coefficients).

Usage

getNodeData(tree, node, dataType = "all")

Arguments

Value

A dataframe which holds all the data that falls into this node of the tree.

Examples

## Not run: 
split_formula <- BMI ~ HISP + WHITE + BLACK + SEX +
  Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, 'ID', nlsySample, degree=1,
  df=3, intercept=TRUE, cp=0.006, minNodeSize=20)

## End(Not run)
node6data <- getNodeData(tree, 6, dataType = 'all')
plot(BMI~AGE, data=node6data)

Sample importance used in vignettes

Description

Sample importance used in vignettes

Usage

importance

Format

An object of class list of length 3.

Get spline coefficients for a single person

Description

Get spline coefficients for a single person

Usage

individual_spline(person, idvar, yvar, tvar, data, boundaryKnots,
  innerKnots, degree, intercept)

Arguments

Baseline socioeconomic information and BMI of 100 individuals.

Description

A dataset containing the body mass index (BMI) and baseline socioeconomic information of 100 individuals from the National Longitudinal Survey of Youth 1979 (NLSY), a freely available longitudinal dataset. The 1000 individuals were drawn randomly from among all NLSY respondents with at least 10 non-missing height/weight responses spread out over at least 20 years. This dataset is used in the package vignettes and code examples. Only a small subset of the variables available from the NLSY are included here. See https://www.bls.gov/nls/nlsy79.htm for more

Usage

nlsySample

Format

A data frame with 16126 rows and 34 columns.

ID

Unique identifier for each NLSY respondent

SEX

Respondent's sex. 1 denotes male, 2 denotes female.

AGE

Respondent's age

BLACK

Indicator for whether or not respondent's identified as Black

BMI

Respondent's body mass index - calculated from reported height and weight

HGC_FATHER

Highest grade completed by respondent's father

HGC_MOTHER

Highest grade completed by respondent's mother

HISP

Indicator for whether or not respondent's race identified as Hispanic

Num_sibs

Number of siblings of respondent

WHITE

Indicator for whether or not respondent identified as white.

HGC

Highest grade completed by respondent

Age_first_smoke

Age that respondent reported first using tobacco. If they reported never using tobacco, recorded as 100.

Age_first_alc

Age that respondent reported first drinking alcohol. If they reported never drinking alcohol, recorded as 100.

RACE

Race, as recorded by NLSY. 1 denotes Hispanic, 2 denotes Black, 3 denotes White.

Source

https://www.bls.gov/nls/nlsy79.htm

Plots the trajectories of each terminal node side by side.

Description

Corresponds to plotting only the second panel of stPlot(). If model$intercept==FALSE, estimated intercepts are added to each trajectory so that the trajectories are plotted at the level of reasonable response values.

Usage

nodePlot(model, colors = NULL)

Arguments

Create a barplot of relative variable importance scores.

Description

Given a named vector of variable importance measures, this function makes a barplot of the relative importances. The importances are scaled to sum to 1. An appropriate input is one column of the output from varImpY() or varImpCoeff().

Usage

plotImp(importance_vector, ...)

Arguments

Examples

imp <- varImpCoeff(forest)[,3]
plotImp(imp, main="Standardized Variable Importance")

Plot the predicted trajectory for a single node

Description

Creates a simple plot of the predicted trajectory at a given node. Option to include the data that falls in the node on the same plot.

Usage

plotNode(tree, node, includeData = FALSE, estimateIntercept = TRUE)

Arguments

Examples

split_formula <- ~HISP + WHITE + BLACK + SEX + Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, idvar = "ID",
   data = nlsySample, degree = 1, df = 3,
   intercept = TRUE, cp = 0.005)

plotNode(tree, 6, includeData=TRUE)

Predict spline coefficients for a testset using a spline tree

Description

Returns a matrix of spline coefficients for each observation in the testset. If no testset is provided, returns predicted coefficients for the individuals in training set; in this case, the columns of the returned predictions correspond to the rows of the flattened training dataset (found in tree$parms$flat_data).

Usage

predictCoeffs(tree, testset = tree$parms$flat_data)

Arguments

Details

importFrom treeClust rpart.predict.leaves

Value

A matrix of spline coefficients. The dimension of the matrix is the degrees of freedom of the spline by the number of units in the test set. The ith column of the matrix holds the predicted coefficients for the ith row in the testset.

Examples

split_formula <- ~HISP + WHITE + BLACK + SEX + Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, idvar = "ID",
   data = nlsySample, degree = 1, df = 3,
   intercept = TRUE, cp = 0.005)

preds <- predictCoeffs(tree)

Predict spline coefficients for a testset using a splineforest.

Description

Uses the forest to predict spline coefficients. Returns a matrix of predicted spline coefficients where the columns of the returned matrix correspond to rows of the testdata. The number of rows of the returned matrix is equal to the degrees of freedom of the forest. If no testdata is provided, forest$flat_data is used. When testdata is not provided, predictions will be made according to one of three methods. The "method" parameter must be either "oob", "itb", or "all". This parameter specifies which trees are used in making a prediction for a certain datapoint. This parameter is not relevant when predicting for a testset that is distinct from the training set.

Usage

predictCoeffsForest(forest, method = "oob", testdata = NULL)

Arguments

Value

A matrix of predicted spline coefficients. The dimensions are forest$df x nrow(testdata). Each column of the matrix corresponds to a row of the testdata.

Examples

trainingSetPreds <- predictCoeffsForest(forest)
newData <- data.frame("WHITE" = 0, "BLACK"=1, "HISP"=0, "Num_sibs"=3,
   "HGC_MOTHER"=12, "HGC_FATHER"=12, "SEX"=1)
predictCoeffsForest(forest, testdata = newData)

Predictions from a spline tree

Description

Returns a vector of predicted responses for the testData. If testData is ommitted, returns predictions for the training data. This function is most meaningful if model$intercept==TRUE.

Usage

predictY(model, testData = NULL)

Arguments

Value

A vector of predictions with rows corresponding to the testdata.

Examples

split_formula <- ~HISP + WHITE + BLACK + SEX + Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, idvar = "ID",
   data = nlsySample, degree = 1, df = 3,
   intercept = TRUE, cp = 0.005)

plot(predictY(tree), tree$parms$data[[tree$parms$yvar]])

Predict responses for a testset using a splineforest.

Description

Uses the forest to make predictions of responses for individuals. This method should only be used on forests where forest$intercept=TRUE. If the testdata parameter is null, makes predictions for each row of the training data. In this case, the methods parameter (which should be set to "oob", "itb", or "all") determines the method used for prediction. If the testdata parameter is not null, the methods parameter is ignored and all trees are used for the prediction of every datapoint.

Usage

predictYForest(forest, method = "oob", testdata = NULL)

Arguments

Value

A vector of predicted responses. The indices of the vector correspond to rows of the testdata.

Examples

trainingSetPreds <- predictYForest(forest)
newData <- data.frame("AGE"=21, "WHITE" = 0, "BLACK"=1, "HISP"=0,
   "Num_sibs"=3, "HGC_MOTHER"=12, "HGC_FATHER"=12, "SEX"=1)
predictYForest(forest, testdata = newData)

Predict responses for the training data

Description

Calling predictY(model) and predict_y_training(model) return identical results, because when no test data is provided to predictY(), the default is to use the training set. This is a slightly faster version that can be used when you know that you wish to predict on the training data. It is faster because it takes advantage of the relationship between model$parms$flat_data and model$parms$data.

Usage

predict_y_training(model)

Arguments

Value

A vector of predicted responses where each element in the vector corresponds to a row in model$parms$data.

Computes percent of variation in projected response explained by a splinetree.

Description

Computes an R^2 measure for a splinetree based on the projected sum of squared errors. Returns 1-SSE/SST. SSE is the sum of projection squared errors between individual smoothed trajectories and predicted smoothed trajectories evaluated on a fixed grid. SST is the sum of projection squared errors between individual smoothed trajectories and the overall population mean trajectory, evaluated on the same fixed grid. If model$intercept==TRUE, then there is the option to ignore the intercept coefficient when computing this metric. When the intercept is ignored, the metric captures how well the model explains variation in shape, and ignores any variation in intercept explained by the model.

Usage

projectedR2(model, includeIntercept = FALSE)

Arguments

Value

The percentage of variation in projected trajectory explained by the model. Computed as 1-SSE/SST. See description.

Examples

r2 <- projectedR2(tree)

Computes a level-based or shape-based evaluation metric for a splineforest.

Description

Computes an R-squared-like evaluation metric for a spline forest. Goal is to see how well the predicted spline coefficients for each individual match the spline coefficients obtained when fitting a spline only to this individual's data (we call these coefficients the true coefficients). Computes 1-SSE/SST, where SSE is the total sum of squared projection errors of the true coefficients compared to the predicted coefficients, and SST is the total sum of squared projection errors of the true coefficients compared to the population mean coefficients. If this is an intercept forest, have the option to compute these sum of squares either with the intercept included or with the intercept ignored to isolate the shape.

Usage

projectedR2Forest(forest, method = "oob", removeIntercept = TRUE)

Arguments

Value

Returns 1-SSE/SST, where SSE is the total sum of squared projection errors of the true coefficients compared to the predicted coefficients, and SST is the total sum of squared projection errors of the true coefficients compared to the population mean coefficients.

Examples

projectedR2Forest(forest, method="all", removeIntercept=TRUE)

projectedR2Forest(forest, method="all", removeIntercept=FALSE)

Prune each tree in forest using a given complexity parameter.

Description

Prunes each tree in the list forest$Trees according to the provided complexity parameter. Returns a new forest.

Usage

pruneForest(forest, cp)

Arguments

Value

A new spline forest model (named list) where each tree has been pruned to the desired level.

Examples

print(avSize(forest))
print(avSize(pruneForest(forest, cp=0.007)))
print(avSize(pruneForest(forest, cp=0.01)))

Calculates coordinates for tree plot

Description

Figures out the coordinates on the tree plot for the each mini trajectory plots. Modified from code from the longRPart package.

Usage

rpartco(tree, parms = paste(".rpart.parms", dev.cur(), sep = "."))

Arguments

Create a faceted spaghetti plot of a splinetree model

Description

Uses ggplot to create a paneled spaghetti plot of the data, where each panel corresponds to a terminal node in the tree. Allows users to visualize homogeneity of trajectories within the terminal nodes of the tree while also looking at the trajectories of different nodes side by side.

Usage

spaghettiPlot(model, colors = NULL)

Arguments

Examples

nlsySubset <- nlsySample[nlsySample$ID %in% sample(unique(nlsySample$ID), 400),]
split_formula <- ~HISP + WHITE + BLACK + SEX + Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, idvar = "ID",
   data = nlsySubset, degree = 1, df = 3,
   intercept = TRUE, cp = 0.005)

spaghettiPlot(tree)

Build a spline random forest.

Description

Builds an ensemble of regression trees for longitudinal or functional data using the spline projection method. The resulting model contains a list of spline trees along with some additional information. All parameters are used in the same way that they are used in the splineTree() function. The additional parameter ntree specifies how many trees should be in the ensemble, and prob controls the probability of selecting a given variable for split consideration at a node. This method may take several minutes to run- saving the forest after building it is recommended.

Usage

splineForest(splitFormula, tformula, idvar, data, knots = NULL,
  df = NULL, degree = 3, intercept = FALSE, nGrid = 7,
  gridPoints = NULL, ntree = 50, prob = 0.3, cp = 0.001,
  minNodeSize = 1, bootstrap = FALSE)

Arguments

Details

The ensemble method is highly similar to the random forest methodology of Breiman (2001). Each tree in the ensemble is fit to a random sample of 63.5 the subset of variables considered at each node is determined by a random process. The prob parameter specifies the probability that a given variable will be selected at a certain node. Because the method is based on probability, the same number of variables are not considered for splitting at each node (as in the randomForest package). Note that if prob is small and the number of variables in the splitFormula is also small, there is a high probability that no variables will be considered for splitting at a certain node, which is problematic. The fewer total variables there are, the larger prob should be to ensure good results.

Value

A spline forest model, which is a named list with 15 components. The list stores a list of trees (in model$Trees), along with information about the spline basis used (model$intercept, model$innerKnots, model$boundaryKnots, etc.), and information about which datapoints were used to build each tree (model$oob_indices and model$index). Note that each element in model$Trees is an rpart object but it is not the same as a model returned from splineTree() because it does not store all relevant information in model$parms.

Examples

nlsySubset <- nlsySample[nlsySample$ID %in% sample(unique(nlsySample$ID), 400),]
splitForm <-~HISP+WHITE+BLACK+HGC_MOTHER+HGC_FATHER+SEX+Num_sibs
sampleForest <- splineForest(splitForm, BMI~AGE, 'ID', nlsySubset, degree=1, cp=0.005, ntree=10)

Build a splinetree model.

Description

Builds a regression tree for longitudinal or functional data using the spline projection method. The underlying tree building process uses the rpart package, and the resulting spline tree is an rpart object with additional stored information. The parameters df, knots, degree, intercept allow for flexibility in customizing the spline basis used for projection. The parameters nGrid and gridPoints allow for flexibility in the grid on which the projection sum of squares is evaluated. The parameters minNodeSize and cp allow for flexibility in controlling the size of the final tree.

Usage

splineTree(splitFormula, tformula, idvar, data, knots = NULL,
  df = NULL, degree = 3, intercept = FALSE, nGrid = 7,
  gridPoints = NULL, minNodeSize = 10, cp = 0.01)

Arguments

Value

An rpart object with additional splinetree-specific information stored in model$parms. The important attributes of the rpart object include model$frame, model$where, and model$cptable. model$frame holds information about each node in the tree. The ith entry in model$where tells us which row of model$frame describes the node that the ith individual in the flattened dataset falls into. model$parms$flat_data holds the flattened dataset that was used to build the tree. model$cptable displays the complexity parameters that would be needed to prune the tree to various desired sizes. Apart from holding the flattened dataset, model$parms holds the boundary knots and the internal knots of the spline basis used to build the tree. These are sometimes important to recover later.

Examples

nlsySample_subset <- nlsySample[nlsySample$ID %in% sample(unique(nlsySample$ID), 500),]
splitForm <- ~HISP+WHITE+BLACK+HGC_MOTHER+HGC_FATHER+SEX+Num_sibs
tree1 <- splineTree(splitForm, BMI~AGE, 'ID', nlsySample_subset, degree=3, intercept=TRUE, cp=0.005)
stPrint(tree1)
stPlot(tree1)

Creates a tree plot of a spline tree.

Description

Creates a tree plot of a spline tree. This corresponds to plotting only the first panel of stPlot(). Code for this function was borrowed from the longRPart package on github.

Usage

splineTreePlot(model, colors = NULL)

Arguments

Custom rpart eval function.

Description

The eval function is required for custom rpart functionality. The split criterion is the total sum of squared errors of the projected or smoothed outcome values around their mean. Note that this is the node purity measure introduced by Yu and Lambert, 1999. The calling of this function is always handled internally by rpart; the user will never directly call this function.

Usage

spline_eval(y, wt = NULL, parms = NULL)

Arguments

Value

A description for the node. This description includes the label, which is the mean response at the node, and the deviance, which in this case is the total projected sum of squares.

Custom rpart init function

Description

The init function is required for custom rpart functionality. This function initializes every node. The init function is responsible for defining the summary function that will be used by rpart's summary function if you call summary() on this tree object. The init function also passes forward its arguments and tells rpart the dimension of the response variable. This function is called internally by rpart; the details are not important for the end user.

Usage

spline_init(y, offset = NULL, parms = NULL, wt = NULL)

Arguments

Value

A list of information for this node that is used internally by rpart.

Custom rpart split function.

Description

The split function is required for the custom rpart functionality. This function is called once per covariate per node during the tree construction, and is responsible for choosing the covariate and threshold for the best split point. This implements the split function suggested by Yu and Lambert. When the covariate is categorical, this code uses a shortcut for computational efficiency. Instead of trying every possible combination of categories as a potential split point, the categories are ordered using the first principal component of the average spline coefficient vector.

Usage

spline_split(y, wt, x, parms = NULL, continuous)

Arguments

Value

A list with two components, goodness and direction, describing the goodness of fit and direction for each possible split for this covariate. The goodness component holds the utility of the split (projected sum of squares) for each possible split. If the continuous parameter is TRUE, goodness and direction each have length n-1, here n is the length of x. The ith value of goodness describes utility of splitting observations 1 to i from i + 1 to n. The values of direction will be -1 and +1, where -1 suggests that values with y < cutpoint be sent to the left side of the tree, and a value of +1 that values with y cutpoint be sent to the right. This is not really an important choice, it only matters for tree reading conventions. If the continuous parameter is FALSE, then the predictor variable x is categorical with k classes and there are potentially almost 2k different ways to split the node. When invoking custom split functions, rpart assumes that a reasonable approximation can be computed by first ordering the groups by their first principal component of the average y vector and then using the usual splitting rule on this ordered variable. In this case, the direction vector has k values giving the ordering of the groups, and the goodness vector has k-1 values giving the utility of the splits.

Custom rpart split function for spline random forests

Description

Wrapper for split function required for the random forest functionality. This function is called once per covariate at each potential split. Implements the random selection of variables; each variable is randomly selected to be included or excluded.

Usage

splineforest_split(y, wt, x, parms = NULL, continuous)

Arguments

Plots a splinetree.

Description

Creates a two paneled plot of a splinetree that shows both the tree and the trajectories side by side. Note that this function has trouble when the plot window is not wide enough. If nothing shows up in RStudio, try increasing the size of the plot window and trying again. For a tree without an intercept, intercepts are estimated after-the-fact for each node using the average starting value in the data so that the plotted trajectories have reasonable response values.

Usage

stPlot(model, colors = NULL)

Arguments

Examples

split_formula <- ~HISP + WHITE + BLACK + SEX + Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, idvar = "ID",
   data = nlsySample, degree = 1, df = 3,
   intercept = TRUE, cp = 0.005)

stPlot(tree, colors = c("red", "orange", "green", "blue", "cyan", "magenta"))

Print a spline tree in the style of print.rpart

Description

The printout provides numbered labels for the terminal nodes, a description of the split at each node, the number of observations found at each node, and the predicted spline coefficients for each node. This code is primarily taken from rpart base code for print.rpart. It has been modified to ensure that the full vector of coefficients is printed for each node.

Usage

stPrint(t, cp, digits = getOption("digits"))

Arguments

Examples

split_formula <- ~HISP + WHITE + BLACK + SEX + Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, idvar = "ID",
   data = nlsySample, degree = 1, df = 3,
   intercept = TRUE, cp = 0.005)

stPrint(tree)

Prints a summary of a terminal node in a tree

Description

If no argument is provided for the parameter node, summaries are printed for every terminal node. Otherwise, the summary of just the requested node is printed.

Usage

terminalNodeSummary(tree, node = NULL)

Arguments

Examples

split_formula <- ~HISP + WHITE + BLACK + SEX + Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, idvar = "ID",
   data = nlsySample, degree = 1, df = 3,
   intercept = TRUE, cp = 0.005)

terminalNodeSummary(tree)

Sample tree used in examples

Description

Sample tree used in examples

Usage

tree

Format

An object of class rpart of length 14.

Given a list of node numbers, returns the depth at which these appear in the tree.

Description

Used in printing and plotting. Source: rpart

Usage

tree.depth(nodes)

Returns a measure of how similar the two trees are.

Description

Computes the Adjusted Rand Index of the clusterings of the population created by the two trees. In the case of correlated covariates, two trees that split on entirely different variables may actually describe similar partitions of the population. This metric allows us to detect when two trees are partitioning the population similarly. A value close to 1 indicates a similar clustering.

Usage

treeSimilarity(tree1, tree2)

Arguments

Value

The Adjusted Rand Index of the clusterings created by the two trees.

See Also

mclust::adjustedRandIndex

Examples

splitForm <- ~SEX+Num_sibs+HGC_MOTHER+HGC_FATHER
nlsySubset <- nlsySample[nlsySample$ID %in% sample(unique(nlsySample$ID), 400),]
tree1 <- splineTree(splitForm, BMI~AGE, "ID", nlsySubset, degree=1, df=2, intercept=FALSE, cp=0.005)
tree2 <- splineTree(splitForm, BMI~AGE, "ID", nlsySubset, degree=1, df=3, intercept=TRUE, cp=0.005)
treeSimilarity(tree1, tree2)

Returns number of terminal nodes in a tree.

Description

Returns number of terminal nodes in a tree.

Usage

treeSize(model)

Arguments

Value

The number of terminal nodes in the tree

Examples

## Not run: 
split_formula <- ~ HISP + WHITE + BLACK + SEX + HGC_FATHER + HGC_MOTHER + Num_sibs
tree <- splineTree(split_formula, BMI~AGE, 'ID', nlsySample, degree=1,
  df=3, intercept=TRUE, cp=0.006, minNodeSize=20)

## End(Not run)
treeSize(tree)

Returns the tree frame.

Description

Provides a similar output to model$frame, but with the redundant information of yval and yval2 removed. Also omits the deviance, the complexity, and the weight. Useful for viewing node numbers and for extracting coefficients for a given node.

Usage

treeSummary(model)

Arguments

Value

A dataframe. The number of rows is the same as the number of nodes in the tree. The row names display the node labels of each node. The "var" attribute either displays the split variable selected at each node, or <leaf> if this node is a terminal node. The "n" attribute displays the number of individuals in the node. The "dev" attribute reports the projected sum of squares at this node; terminal nodes have the smallest values for "dev" because this is what the tree building process is supposed to minimize. The "coeffs" attribute displays the coefficients predicted for each node.

Examples

nlsySubset <- nlsySample[nlsySample$ID %in% sample(unique(nlsySample$ID), 400),]
split_formula <- ~HISP + WHITE + BLACK + SEX + Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, idvar = "ID",
   data = nlsySubset, degree = 1, df = 3,
   intercept = TRUE, cp = 0.005)

treeSummary(tree)

Random Forest Variable Importance based on spline coefficients

Description

Returns the random forest variable importance based on the permutation accuracy measure, which is calculated as the difference in mean squared error between the original data and from randomly permuting the values of a variable.

Usage

varImpCoeff(forest, removeIntercept = TRUE, method = "oob")

Arguments

Value

a matrix of variable importance metrics.

Examples

importanceMatrix <- varImpCoeff(forest, removeIntercept=TRUE)

Random Forest Variable Importance based on Y

Description

Returns the random forest variable importance based on the permutation accuracy measure, which is calculated as the difference in mean squared error between the original data and from randomly permuting the values of a variable.

Usage

varImpY(forest, method = "oob")

Arguments

Details

The "method" parameter deals with the way in which forest performance should be measured. Since variable importance is based on a change in performance, the "method" parameter is necessary for a variable importance measure. The choices are "oob" (out of bag), "all", or "itb" (in the bag).

Value

A matrix storing variable importance metrics. The rows correspond to split variables. The columns are different methods of measuring importance. The first column is the absolute importance (mean difference in performance between permuted and unpermuted datasets). The second column measures the mean percent difference in performance. The third column standardizes the differences by dividing them by their standard deviation.

Examples

importanceMatrix <- varImpY(forest, method="oob")
plotImp(importanceMatrix[,3])

Computes percent of variation in response explained by spline tree.

Description

Computes the percentage of variation in response explained by the spline tree. This metric is only meaningful if model$intercept==TRUE. If the tree includes an intercept, the measure will be between 0 and 1.

Usage

yR2(model)

Arguments

Value

An R^2 goodness measure. 1-SSE/SST where SSE is the sum of squared errors between predicted responses and true responses, and SST is sum of squared errors of true responses around population mean. Note that if the tree passed in was built without an intercept, this function will return NULL.

Examples

split_formula <- ~HISP + WHITE + BLACK + SEX + Num_sibs + HGC_FATHER + HGC_MOTHER
tree <- splineTree(split_formula, BMI~AGE, idvar = "ID",
   data = nlsySample, degree = 1, df = 3,
   intercept = TRUE, cp = 0.005)

yR2(tree)

Computes a level-based evaluation metric for a splineforest that was built WITH an intercept.

Description

Computes the R-squared metric for a spline forest. Goal is to see how well the predicted response values match the actual response values. Note that this function should only be used on forests where the intercept parameter is TRUE. A simple 1-SSE/SST calculation.

Usage

yR2Forest(forest, method = "oob")

Arguments

Value

Returns 1-SSE/SST, where SSE is the total sum of squared errors of the true responses and predicted responses, and SST is the total sum of squared errors of the responses around their mean. If this forest was not built with an intercept, returns NULL.

Examples

yR2Forest(forest, method="all")