fit: Fitting a multivariate process model

In smde: Sparse Multivariate Differential Equations

Description

Estimation of parameters in a multivariate process model. The function fits the models by minimizing an l1-penalized squared error loss by a coordinate wise descent algorithm. Alternatively, the function can do unpenalized least squares forward or backward model search.

Usage

fit(model, ...)

## S3 method for class 'multModel'
fit(model, method = "cd", scope = seq(1, model$p), ...)

Arguments

model
method	a character. The default, "cd", means a standard coordinate descent algorithm. while 'mf' means the same coordinate descent algorithm, but using a matrix free implementation. A choice of 'forward' or 'backward' means least squares model search.
scope	a numeric or a list. See stepOptim.
...	additional arguments passed on to the optimization algorithm.

Details

The stepwise model search algorithms as well as the algorithm for l1-panalized estimation return a sequence of models parametrized by a vector of tuning parameters. For the model search, the tuning parameter is the dimension of the model (the number of nonzero entries in the parameter vector). For the l1-panalized models, the tuning parameter is a sequence of penalty parameters.

The scope argument specifies the smallest and largest model fitted. They must be nested. The zero weights of the multModel object are used to specify the smallest model, if the scope argument does not contain a smallest model.

Value

a multModel object with the additional elements

lambda	the sequence of tuning parameters
Blambda	the corresponding matrix of estimated parameters. The i'th column corresponds to the i'th tuning parameter.
status	either 0 (meaning no errors), 1 (convergence), 2 (MLE not computed) or 3 (other errors) indicating different problems or errors
msg	errors encountered or other notable conditions

See Also

multModel, coordinateDescent, coordinateDescentMF, stepOptim.

smde documentation built on May 2, 2019, 4:58 p.m.