makeDataTypes {Umpire}

Discretize a Continuous Data Set to Mixed Types

Description

The makeDataTypes function allows the user to discretize a continuous data set generated from an engine of any type into binary, nominal, ordinal, or mixed-type data.

Usage

makeDataTypes(dataset, pCont, pBin, pCat, pNominal = 0.5,
             range = c(3, 9), exact = FALSE, inputRowsAreFeatures = TRUE)
getDataTypes(object)
getDaisyTypes(object)

Arguments

dataset

A matrix or dataframe of continuous data.

pCont

Non-negative, numeric probability equal to or between 0 and 1 describing desired percent frequency of continuous features.

pBin

Non-negative, numeric probability equal to or between 0 and 1 describing desired percent frequency of binary features.

pCat

Non-negative, numeric probability equal to or between 0 and 1 describing desired percent frequency of categorical features.

pNominal

Non-negative, numeric probability equal to or between 0 and 1 describing desired percent frequency of categorical features to be simulated as nominal.

range

A set of integers whose minimum and maximum determine the range of the number of levels to be associated with simulated categorical factors.

exact

A logical value; should the parameters pCont, pBin and pCat be treated as exact fractions to achieve or as probabilities.

inputRowsAreFeatures

Logical value indicating if features are to be simulated as rows (TRUE) or columns (FALSE).

object

An object of the MixedTypeEngine class.

Details

The makeDataTypes function is a critical step in the construction of a MixedTypeEngine, which is an object that is used to simulate clinical mixed-type data instead of gene expression data. The standard process, as illustrated in the example, involves (1) creating a ClinicalEngine, (2) generating a random data set from that engine, (3) adding noise, (4) setting the data types, and finally (5) creating the MixedTypeEngine.

The main types of data (continuous, binary, or categorical) are randomly assigned using the probability parameters pCont, pBin, and pCat. To choose the splitting point for the binary features, we compute the bimodality index (Wang et al.). If that is significant, we split the data halfway between the two modes. Otherwise, we choose a random split point between 5% and 35%. Categorical data is also randomly assigned to ordinal or nominal based on the pNominal parameter. The number of levels is uniformly selected in the specified range, and the fraction of data points assigned to each level is selected from a Dirichlet distribution with alpha = c(20, 20, ..., 20).

Value

A list containing:

binned

An object of class data.frame of simulated, mixed-type data.

cutpoints

An object of class "list". For each feature, data type, labels (if data are binary or categorical), and break points (if data are binary or categorical) are stored. Cutpoints can be passed to a MixedTypeEngine for storage and downstream use.

The getDataTypes function returns a character vector listing the type of data for each feature in a MixedTypeEngine. the getDaisyTypes function converts this vector to a list suitable for input to the daisy function in the cluster package.

Note

If pCont, pBin, and pCat do not sum to 1, they will be rescaled.

By default, categorical variables are simulated as a mixture of nominal and ordinal data. The remaining categorical features described by pNominal are simulated as ordinal.

Author(s)

Kevin R. Coombes [e-mail], Caitlin E. Coombes [e-mail]

References

Wang J, Wen S, Symmans WF, Pusztai L, Coombes KR.
The bimodality index: A criterion for discovering and ranking bimodal signatures from cancer gene expression profiling data.
Cancer Inform 2009; 7:199-216.

See Also

MixedTypeEngine

Examples

## Create a ClinicalEngine with 4 clusters
ce <- ClinicalEngine(20, 4, TRUE)

## Generate continuous data
set.seed(194718)
dset <- rand(ce, 300)

## Add noise before binning mixed type data
cnm <- ClinicalNoiseModel(nrow(ce@localenv$eng)) # default
noisy <- blur(cnm, dset$data)

## Set the data mixture
dt <- makeDataTypes(dset$data, 1/3, 1/3, 1/3, 0.3)
cp <- dt$cutpoints
type <- sapply(cp, function(X) { X$Type })
table(type)
summary(dt$binned)

## Use the pieces from above to create an MTE.
mte <- MixedTypeEngine(ce,
           noise = cnm,
           cutpoints = dt$cutpoints)
## and generate some data with the same data types and cutpoints
R <- rand(mte, 20)
summary(R)