mfFamily: Manifold Gradient Boosting Families

In Almond-S/manifoldboost: Boosting Regression Models for Manifold Valued Data

Manifold Gradient Boosting Families

Description

mfboost_family objects extend boost_family objects to contain, i.a., a geometry slot, which is used to transfer geometric structure information of the response into the functions defined in a boost_family. It is used to specify loss, risk and other functions for fitting regression problems with manifold valued response with mfboost in a flexible and modular way.

Usage

mfFamily(
  mf,
  pole,
  ngradient,
  update_formula = function(timeformula, pole_) timeformula,
  loss = NULL,
  risk = NULL,
  response = function(f) NA,
  nuisance = function() return(NA),
  weights = c("any", "none", "zeroone", "case"),
  name = "user-specified mfFamily",
  offset = function(y, w) 0,
  fW = NULL
)

RiemannL2(
  mf,
  pole.type = c("RiemannL2", "Gaussian"),
  pole.control = boost_control()
)

PlanarShapeL2(
  pole.type = "RiemannL2",
  pole.control = boost_control(),
  weight_fun = equal_weights,
  arg_range = NULL
)

PlanarSizeShapeL2(
  pole.type = "RiemannL2",
  pole.control = boost_control(),
  weight_fun = equal_weights,
  arg_range = NULL
)

EuclideanL2(
  cyclic = FALSE,
  arg.grid.len = 5000,
  weights = weights,
  arg.range = NULL,
  smoothed.cov = NULL,
  cov.k = 10
)

Arguments

mf	The response geometry supplied as an mfGeometry object.
pole	a function with arguments formula and data (in a format used by FDboost) returning the pole in the format required by mf.
ngradient, loss, risk, response, nuisance	functions defining the mboost::boost_family object.
update_formula	a function with arguments formula and pole_ updating the formula in dependence of the pole_, typically by applying a function to the right side.
weights, name	boost_family character strings indicating the weights allowed and the name of the loss function.
offset, fW, rclass, check_y, type	other slots of boost_family objects that typically do not play a role in mfboost_family objects.
pole.type	one of "RiemannL2"(default) and "Gaussian"
pole.control	a list of parameters controlling the mboost algorithm. For more details see boost_control.
weight_fun	a function producing inner product weights taking the arguments arg (vector of arguments of the function) and range (range of the arguments). Passed to mf$initialize.
arg_range	vector of length 2 specifying the range argument of the weight_fun. The default NULL will take the minimum and maximum of the supplied arg.
initialize	a function with arguments formula (in obj.formula format) and data (in a format used by FDboost) initializing mf by setting mf$y_ and mf$weights_.

Almond-S/manifoldboost documentation built on June 23, 2022, 11:06 a.m.