Model Initialization
LRMoE.cmm_init
— Functioncmm_init(Y, X, n_comp, type; exact_Y = false, n_random = 5)
Initialize an LRMoE model using the Clustered Method of Moments (CMM).
Arguments
Y
: A matrix of response.X
: A matrix of covariates.n_comp
: Integer. Number of latent classes/components.type
: A vector of eithercontinuous
,discrete
orreal
, indicating the type of response by dimension.
Optional Arguments
exact_Y
:true
orfalse
(default), indicating ifY
is observed exactly or with censoring and truncation.n_random
: Integer. Number of randomized initializations.
Return Values
zero_y
: Proportion of zeros in observedY
.mean_y_pos
: Mean of positive observations inY
.var_y_pos
: Variance of positive observations inY
.skewness_y_pos
: Skewness of positive observations inY
.kurtosis_y_pos
: Kurtosis of positive observations inY
.α_init
: Initialization of logit regression coefficientsα
.params_init
: Initializations of expert functions. It is a three-dimensional vector. For example,params_init[1][2]
initializes the 1st dimension ofY
using the 2nd latent class, which is a vector of potential expert functions to choose from.ll_init
: Calculates the loglikelihood of each expert function on the clustered groups ofY
. For example,ll_init[1][2][3]
is the loglikelihood of the 1st dimension ofY
, calculated based on the 2nd latent classes and the 3rd initialized expert function inparams_init
.ll_best
: An initialization chosen fromparams_init
which yields the highest likelihood upon initialization.random_init
: A list ofn_random
randomized initializations chosen fromparams_init
.