predict.MoE_gating {MoEClust}

Predictions from MoEClust gating networks

Description

Predicts mixing proportions from MoEClust gating networks. Effectively akin to predicting from a multinomial logistic regression via multinom, although here the noise component (if any) is properly accounted for. So too are models with no gating covariates at all, or models with the equal mixing proportion constraint. Prior probabilities are returned by default.

Usage

## S3 method for class 'MoE_gating'
predict(object,
        newdata = NULL,
        type = c("probs", "class"),
        keep.noise = TRUE,
        droplevels = FALSE,
        ...)

## S3 method for class 'MoE_gating'
fitted(object,
       ...)

## S3 method for class 'MoE_gating'
residuals(object,
          ...)

Arguments

object

An object of class "MoE_gating" (typically x$gating, where x is of class "MoEClust").

newdata

A matrix or data frame of test examples. If omitted, the fitted values are used.

type

The type of output desired. The default ("probs") returns prior probabilities, while "class" returns labels indicating the most likely group a priori. Note that observations classified assigned the noise component (if any) are given a label of 0.

keep.noise

A logical indicating whether the output should acknowledge the noise component (if any). Defaults to TRUE; when FALSE, this column is discarded and the matrix of probabilities is renormalised accordingly.

droplevels

A logical indicating whether unseen factor levels in categorical variables within newdata should be dropped (with NA predicted in their place). Defaults to FALSE. See drop_levels.

...

Catches unused arguments or allows the type and keep.noise arguments to be passed through fitted and the keep.noise argument to be passed through residuals.

Details

This function is effectively a shortcut to predict.MoEClust(x, ...)$pro, which (unlike the predict method for multinom on which predict.MoE_gating is based) accounts for the various ways of treating gating covariates, equal mixing proportion constraints, and noise components, although its type argument defaults to "probs" rather than "class".

Value

The return value depends on whether newdata is supplied or not and whether the model includes gating covariates to begin with. When newdata is not supplied, the fitted values are returned (as a matrix if the model contained gating covariates, otherwise as a vector as per x$parameters$pro). If newdata is supplied, the output is always a matrix with the same number of rows as the newdata.

Note

Note that the keep.noise argument does not correspond in any way to the discard.noise argument to predict.MoEClust; there, the noise component is respected in the computation of the mixing proportions and only discarded (if at all) in the prediction of the responses.

Author(s)

Keefe Murphy - <[e-mail]>

References

Murphy, K. and Murphy, T. B. (2020). Gaussian parsimonious clustering models with covariates and a noise component. Advances in Data Analysis and Classification, 14(2): 293-325. <doi:10.1007/s11634-019-00373-8>.

See Also

predict.MoEClust, multinom, predict.MoE_expert, drop_levels

Examples

data(ais)
mod    <- MoE_clust(ais[,3:7], G=2, modelNames="EEE", gating= ~ SSF + Ht,
                 expert= ~ sex, network.data=ais, tau0=0.1, noise.gate=FALSE)
(preds <- predict(mod$gating, newdata=ais[1:5,]))

all.equal(preds, predict(mod, newdata=ais[1:5,])$pro) #TRUE

# Note that the predictions are not the same as the multinom predict method
# in this instance, owing to the invocation of noise.gate=FALSE above
mod2   <- mod 
class(mod2$gating) <- c("multinom", "nnet")
predict(mod2$gating, newdata=ais[1:5,], type="probs")

# We can make this function behave in the same way by invoking keep.noise=FALSE
predict(mod$gating, keep.noise=FALSE, newdata=ais[1:5,])

# ... although keep.noise=FALSE in predict.MoE_gating does not
# yield the same output as discard.noise=TRUE in predict.MoEClust
predict(mod, discard.noise=TRUE, newdata=ais[1:5,])$pro