online_log_mixture: Create an S4 object to fit the online logistic mixture model

In MKaptein/ofmlr: Fit online finite mixtures of logistic regression models

Description

This function allows you to initialize an online logistic regression model. After initialization by providing the model description (in terms of the number of predictors $p$ of each logistic regression model and the number of mixture components $k$) one can use the add.observation method to add observations and update the parameters and call the summary() and plot() methods.

Usage

online_log_mixture(p, k, beta = matrix(runif(k * p, -1, 1), nrow = k),
  ak = generate_probability_vector(k), ll.window = 500, trace = FALSE)

Arguments

p	Number of predictors for each logistic regression model
k	Number of mixture components
beta	A $k$ by $p$ matrix with the true regression coefficients for each mixture component
ak	A vector of length $k$ summing to 1 for the mixture probabilties
ll.window	The length of the (lagged) mean log-likelihood window
trace	Whether or not the parameter estimates and the lagged mean log likelihood should be traced can be set to an integer, $x > 0$, and it will store a snapshot every x-th datapoint. WARINING: when using trace=1 the object grows large very quickly

Value

An object of class "online_log_mixture" The object contains the following slots:

params

A list containing all the parameters of the model. Contains the objects beta, Sak, n

descriptives

A list containing the (lagged) mean log-likelihood, the max log-likelihood, and the AIC and BIC approximations

trace

A list containing the trace of each of the parameters

See Also

For more detailed examples see add_observation

Examples

online_log_mixture(3,2)
online_log_mixture(2,4, ak=c(.1,.2,.3,.4), trace=100)

MKaptein/ofmlr documentation built on Sept. 25, 2020, 11:46 a.m.