mixture (version 2.1.1)

e_step: Expectation Step

Description

Calculates the expectation of class memberships, and imputes if missing values for a given dataset.

Usage

e_step(data, model_obj, start=0, nu = 1.0)

Value

Returns a list with the following components:

X

A matrix of the original dataset plus imputed values if applicable.

origX

A matrix of the original dataset including missing values.

map

A vector of integers indicating the maximum a posteriori classifications for the best model.

z

A matrix giving the raw values upon which map is based.

row_tags

If there were NAs in the original dataset, a vector of indices referencing the row of the imputed vectors is given.

Arguments

data

A matrix or data frame such that rows correspond to observations and columns correspond to variables. Note that this function currently only works with multivariate data p > 1.

start

Start values in this context are only used for imputation. Non-missing values have their expectation of class memberships calculated directly. If 0 then the random soft function is used for initialization. If 1 then the random hard function is used for initialization. If 2 then the kmeans function is used for initialization. If is.matrix then matrix is used as an initialization matrix as along as it has non-negative elements. Note: only models with the same number of columns of this matrix will be fit.

model_obj

A gpcm_best, vgpcm_best, stpcm_best, ghpcm_best, and salpcm_best object class.

nu

deterministic annealing for the class membership E-step.

Author

Nik Pocuca, Ryan P. Browne and Paul D. McNicholas.

Maintainer: Paul D. McNicholas <[e-mail]>

Details

This will only work on a dataset with the same dimension as estimated in the model. e_step will also work for missing values, provided that there is at least one non-missing entry.

References

Browne, R.P. and McNicholas, P.D. (2014). Estimating common principal components in high dimensions. Advances in Data Analysis and Classification 8(2), 217-226.

Zhou, H. and Lange, K. (2010). On the bumpy road to the dominant mode. Scandinavian Journal of Statistics 37, 612-631.

Celeux, G., Govaert, G. (1995). Gaussian parsimonious clustering models. Pattern Recognition 28(5), 781-793.

Examples

if (FALSE) {
# load dataset and perform model search. 

data(x2)
data_in <- matrix(x2,ncol = 2)
mm <- mixture::gpcm(data = data_in,G = 1:7,
           start = 0,
           veo = FALSE,pprogress=FALSE)

# get best model 
best = get_best_model(mm)
best 

# lets try imputing some missing data. 
x2NA <- x2
x2NA[5,1] <- NA
x2NA[140,2] <- NA
x2NA[99,1] <- NA

# calculate expectation
expect <- e_step(data=x2NA,start = 0,nu = 1.0,model_obj = best)

# plot imputed entries and compare with original 
plot(x2,col = "grey")
points(expect$X[expect$row_tags+1,],col = "blue", pch = 20,cex = 2) # blue are imputed values.
points(x2[expect$row_tags+1,], col = "red" , pch = 20,cex = 2) # red are original values.
legend(-2,2,legend = c("imputed","original"),col = c("blue","red"),pch = 20)
}