initialize: Intialization functions for plvm models

In bozenne/lava.penalty: Penalized Latent Variable Models

Description

initializeFactor.penaltyL12: update the lasso penalty to group lasso penalty in presence of categorical variables. initialize.penaltyL12: estimate starting value for the plvm. initialize.penaltyNuclear: estimate starting value for the plvm. initialize.start: estimate starting value for the plvm.

Usage

initializeFactor.penaltyL12(x, data, trace)

initialize.penaltyL12(x, name.x, model.x, regularizationPath)

initialize.start(x, data, regularizationPath, increasing, constrain.variance,
  name.variance, ...)

initializer.penaltyNuclear(x, name.coef)

Arguments

x	a penalized lvm model
data	a data.frame containing the data
trace	should the user be told that some penalties have been updated ?
name.coef	the name of the parameters

Details

Remove penalties corresponding to reference links

Examples

#### initialize.data ####


#### initializeFactor.penaltyL12 ###
m <- lvm(Y~X1+X2+X3+X4)
pm <- penalize(m)
initializeFactor.penaltyL12(pm, data = sim(m, 1), trace = TRUE)

mCAT <- m
categorical(mCAT, K = 3, labels = letters[1:3]) <- "X1"
categorical(mCAT, K = 2, labels = letters[1:2]) <- "X2"

pm <- penalize(m)
initializeFactor.penaltyL12(pm, data = sim(mCAT, 1), trace = TRUE)

pmCAT <- penalize(mCAT)
initializeFactor.penaltyL12(pmCAT, data = sim(mCAT, 1), trace = TRUE)

#### initialize.penaltyL12 #### 
m <- lvm()
regression(m) <- y1~x1+x2+x3+x4

#### elastic net
e <- estimate(m, sim(m, 1e2))
pm <- penalize(m, lambda1 = 2, lambda2 = 1.5)
pen12 <- initialize.penaltyL12(pm$penalty, name.coef = names(coef(e)))
pen12

i <- initializer.plvm(pm, data = sim(m, 1e2),
                 regularizationPath = FALSE, constrain.variance = FALSE)

#### group lasso
categorical(m, labels = c("A","B","C")) <- "x1"
categorical(m, labels = c("A","B","C")) <- "x2"

e <- estimate(m, sim(m, 1e2))
pm <- penalize(m, lambdaG = 5)
pen12 <- initializer.penaltyL12(pm$penalty, name.coef = names(coef(e)))
pen12
estimate(pm, data = sim(pm, 1e2))

i <- initializer.plvm(pm, data = sim(m, 1e2),
                 regularizationPath = FALSE, constrain.variance = FALSE)

# nuclear norm
m <- lvm()
coords <- expand.grid(x = 1:10, y = 1:10)
m <- regression(m, y = "y1", x = paste0("z",1:100))

mNuclear <- lvm(y1 ~ x1 + x2)
penalizeNuclear.penaltyL12(mNuclear, coords = coords, lambdaN = 10) <- coefReg(m, value = TRUE)
mNuclear

penN <- initializer.plvm(mNuclear$penaltyNuclear, name.coef = coef(mNuclear))
penN

bozenne/lava.penalty documentation built on May 13, 2019, 1:41 a.m.