autocartControl {autocart} | R Documentation |
Create the object used for the controlling of the splits in the autocart model
Description
Create the object used for the controlling of the splits in the autocart model
Usage
autocartControl(
minsplit = 20,
minbucket = round(minsplit/3),
maxdepth = 30,
maxobsMtxCalc = NULL,
distpower = 1,
islonglat = TRUE,
givePredAsFactor = TRUE,
retainCoords = TRUE,
useGearyC = FALSE,
runParallel = TRUE,
spatialWeightsType = "default",
customSpatialWeights = NULL,
spatialBandwidthProportion = 1,
spatialBandwidth = NULL,
asForest = FALSE
)
Arguments
minsplit |
The minimum observations in a node before a split is attempted |
minbucket |
The minimum number of observations in a terminal node. |
maxdepth |
Set the maximum depth in the final tree. A root node is counted as a height of 0. |
maxobsMtxCalc |
Optional maximum number of observations in a node where computationally intensive matrix calculations like autocorrelation and compactness are performed. |
distpower |
The power of inverse distance to use when calculating spatial weights matrix. |
islonglat |
Are the coordinates longitude and latitude coordinates? If TRUE, then use great circle distance calculations |
givePredAsFactor |
In the returned autocart model, should the prediction vector also be returned as a factor? |
retainCoords |
After creating the autocart model, should the coordinates for each of the predictions be kept in the returned model? |
useGearyC |
Should autocart use Geary's C instead of Moran's I in the splitting function? |
runParallel |
Logical value indicating whether autocart should run using parallel processing. |
spatialWeightsType |
What type of spatial weighting should be used when calculating spatial autocorrelation? Options are "default" or "gaussian". |
customSpatialWeights |
Use this parameter to pass in an optional spatial weights matrix for use in autocorrelation calculations. Must have nrow and ncol equal to rows in training dataframe. |
spatialBandwidthProportion |
What percentage of the maximum pairwise distances should be considered the maximum distance for spatial influence? Cannot be simultaneously set with |
spatialBandwidth |
What is the maximum distance where spatial influence can be assumed? Cannot be simultaneously set with |
asForest |
A logical indicating if the tree should be created as a forest component with random subsetting of predictors at each node. Set this to true if you are using this tree inside an ensemble. |
Value
An object passed in to the autocart
function that controls the splitting.
Examples
# Load some data for an autocartControl example
snow <- na.omit(read.csv(system.file("extdata", "ut2017_snow.csv", package = "autocart")))
y <- snow$yr50[1:40]
X <- data.frame(snow$ELEVATION, snow$MCMT, snow$PPTWT)[1:40, ]
locations <- as.matrix(cbind(snow$LONGITUDE, snow$LATITUDE))[1:40, ]
# Create a control object that disallows the tree from having a depth more
# than 10 and give spatial weights only to observations that are a third of the
# distance of the longest distance between any two points in the dataset.
snow_control <- autocartControl(maxdepth = 10, spatialBandwidthProportion = 0.33)
# Pass the created control object to an autocart model
snow_model <- autocart(y, X, locations, 0.30, 0, snow_control)