mmpcaClust-class: mmpcaClust class

Description Details Slots Objects from the class

Description

An S4 class representing a fitted mmpca model.

Details

The BB-CVEM method is the branch & bound greedy procedure proposed in the original paper of Jouvin et. al. https://arxiv.org/abs/1909.00721. The number of epochs in the n_epochs slot is actually the true number of pass minus 1 (unless max.epochs was reached). Indeed, the last pass before convergence does not change either the bound or the clustering, hence it is removed of the counter.

Slots

call

A call object specifying the call

method

The method used in the call

clustering

The final partition found by the algorithm

controls

An object of class mmpcaClustcontrol containing the controls used in the VEM algorithm on the aggregated DTM during the loop. The slots controls@control_lda_init where only use when init.beta == 'lda'.

K

An integer specifying the number of topics.

Q

An integer specifying he number of clusters.

N

An integer specifying the number of observations.

V

An integer specifying the number of variables.

beta

The (KxV) topic matrix.

gamma

A (QxK) matrix containing the variational paramaters of the variational distribution of each $\theta_q$ in its rows.

lda_algo

An object of class "LDA" (cf. TopicModel) containing the results of the LDA() function applied to the aggregated DTM, with control controls@control_lda_loop

max.epochs

The maximum number of pass through the whole dataset in the algorithm.

logLikelihoods

A numeric vector containing the evolution of the variational bound every keep iteration.

keep

An integer specifying the . Mostly useful for the plot function.

n_epochs

The number of pass through the datasets before convergence. see details

llhood

The final value of the variational lower bound.

Yinit

The value of the initial partition.

icl

The Integrated Classification Likelihood value.

Objects from the class

Object of class "mmpcaClust" are returned by mmpca_clust()


MoMPCA documentation built on July 1, 2020, 9:36 p.m.