This is a wrapper for the stand-alone version of R's earth package.

Pkg.clone("git@github.com:lendle/Earth.jl.git")
Pkg.build("Earth")

Ijulia notebook example

Returns an object of type EarthFit

Arguments

• x::VecOrMat{Float64} - Vector or matrix of predictors. size(x, 1) is number of observations. size(x, 2) is number of predictors.
• y::VecOrMat{Float64} - Vector or matrix of outcomes. size(y, 1) is number of observations. size(y, 2) is number of outcomes.
• WeightsArg = ones(size(x, 1)) - Vector of observation weights. Comments in the C code indicate that this may not do anything.
• nMaxDegree = 1 - Maximum degree of interactions.
• nMaxTerms = min(200, max(20, 2 * size(x, 2))) + 1 - Maximum number of model terms before pruning.
• Penalty = (nMaxDegree > 1)? 3.0: 2.0 - Generalized Cross Validation (GCV) penalty per knot.
• Thresh = 0.001 - Forward stepping threshold.
• nMinSpan = 0 - Minimum distance between knots.
• Prune = true - Perform backwards pass?
• nFastK = 20 - Maximum number of parent terms considered at each step of the forward pass.
• FastBeta = 0.0 - Fast MARS aging coefficient, as described in the Fast MARS paper section 3.1.
• NewVarPenalty = 0.0 - Penalty for adding a new variable in the forward pass.
• LinPreds=zeros(Int32, size(x, 2)) - Index vector specifying which predictors should enter linearly, as in linear regression. This does not say that a predictor must enter the model; only that if it enters, it enters linearly.
• UseBetaCache = false - Using the “beta cache” takes more memory but is faster (by 20% and often much more for large models).
• Trace = 0.0 - Set to higher values (up to 5.0?) for increasingly verbose output. Set to special value 1.5 for information about memory allocations.

See the documentation and vignette from the R package for details on the optional arguments, which may have slightly different names.

Returns a matrix of predicted outcomes with one row per observation and one column per outcome variable.

Arguments

• ef::EarthFit - fitted earth object.
• x::VecOrMat{Float64} - Vector or matrix of (possibly new) predictors.

Given a vector or matrix of (possibly new) predictors, returns a matrix of MARS terms with a constant term in the first column.

Arguments

• ef::EarthFit - fitted earth object.
• x::VecOrMat{Float64} - Vector or matrix of (possibly new) predictors.