mlt (version 1.7-4)

mlt: Most Likely Transformations

Description

Likelihood-based model estimation in conditional transformation models

Usage

mlt(model, data, weights = NULL, offset = NULL, fixed = NULL,
    theta = NULL, pstart = NULL, scaleparm = TRUE,
    dofit = TRUE, optim = mltoptim(hessian = has_scale(model)))

Value

An object of class mlt with corresponding methods.

Arguments

model

a conditional transformation model as specified by ctm

data

a data.frame containing all variables specified in model

weights

an optional vector of case weights

offset

an optional vector of offset values; offsets are not added to an optional scaling term (see link{ctm})

fixed

a named vector of fixed regression coefficients; the names need to correspond to column names of the design matrix

theta

optional starting values for the model parameters

pstart

optional starting values for the distribution function evaluated at the data

scaleparm

a logical indicating if (internal) scaling shall be applied to the model parameters; TRUE unless a location-scale model is fitted where the numerically approximated Hessian is only available on the original parameter scale

dofit

a logical indicating if the model shall be fitted to the data (TRUE) or not. If theta is given, a model of class mlt (a full "fitted" model) featuring these parameters is returned. Otherwise, an unfitted model of class ctm is returned

optim

a list of functions implementing suitable optimisers, requires numerical Hessian for location-scale models

Details

This function fits a conditional transformation model by searching for the most likely transformation as described in mlt::Hothorn:Moest:Buehlmann:2017, including location-scale models Siegfried_Kook_Hothorn_2023. Implementation details are given in mlt::Hothorn:2018.

References

*

Examples

 
  ### set-up conditional transformation model for conditional
  ### distribution of dist given speed
  dist <- numeric_var("dist", support = c(2.0, 100), bounds = c(0, Inf))
  speed <- numeric_var("speed", support = c(5.0, 23), bounds = c(0, Inf)) 
  ctmm <- ctm(response = Bernstein_basis(dist, order = 4, ui = "increasing"),
              interacting = Bernstein_basis(speed, order = 3))

  ### fit model
  mltm <- mlt(ctmm, data = cars)

  ### plot data
  plot(cars)
  ### predict quantiles and overlay data with model via a "quantile sheet"
  q <- predict(mltm, newdata = data.frame(speed = 0:24), type = "quantile", 
               p = 2:8 / 10, K = 500)
  tmp <- apply(q, 1, function(x) lines(0:24, x, type = "l"))