Data format used in RMixtComp

Quentin Grimonprez

2020-01-07

Parameters

mixtCompLearn requires 3 R objects: algo (a list), data (a list, a data.frame or a matrix) and model (a list).

Data

Data must have one of the following format. Each variable must be named.

data <- list(
  varName1 = c(elem11, elem12, elem13, elem14),
  varName2 = c(elem21, elem22, elem23, elem24),
  varName3 = c(elem31, elem32, elem33, elem34),
)

data <- data.frame(
  varName1 = c(elem11, elem12, elem13, elem14),
  varName2 = c(elem21, elem22, elem23, elem24),
  varName3 = c(elem31, elem32, elem33, elem34),
)

data <- matrix(c(elem11, elem12, elem13, elem14,
                 elem21, elem22, elem23, elem24,
                 elem31, elem32, elem33, elem34), 
                 ncol = 3, dimnames = list(NULL, c("varName1", "varName2", "varName3")))

Exemple:

Model

model is a list describing the variables used for clustering and the distribution used. Each element corresponds to a variable and contains two elements: the model used (type), and the hyperparameters of the model if any (paramStr). When there is no hyperparameters, instead of a list, user can just provide the model name.

model <- list(varName1 = list(type = "Model1", paramStr = "param1"),
              varName2 = "Model2",
              varName3 = list(type = "Model3", paramStr = ""))

The model object can contain less variables than the data object. Only variables listed in the model object are used for clustering.

Exemple:

Algo

algo is a list containing the required parameters of the SEM algorithm.

User can add extra elements, they will be copied in the output object.

To easily create this list, the function createAlgo can be used. It creates the desired list with default values.

Available Models

Eight models are available in RMixtComp

Overview

Available models Data type Restrictions Hyperparameters
Gaussian Real
Weibull Real >= 0
Poisson Integer >= 0
NegativeBinomial Integer >= 0
Multinomial Categorical
Rank_ISR Rank
Func_CS Functional yes
Func_SharedAlpha_CS Functional yes

Details

Gaussian

For real data. For a class \(k\), parameters are the mean (\(\mu_k\)) and the standard deviation (\(\sigma_k\)). The distribution function is defined by:

\[ f_k(x) = \frac{1}{\sqrt{2\pi\sigma_k^2}}\exp{\left(-2\frac{(x-\mu_k)^2}{\sigma_k^2}\right)} \]

Weibull

For positive real data (usually lifetime). For a class \(j\), parameters are the shape \(k_j\) and the scale \(\lambda_j\). The distribution function is defined by:

\[ f_j(x) = \frac{k_j}{\lambda_j} \left(\frac{x}{\lambda_j}\right)^{k_j-1} \exp{\left(-\left(\frac{x}{\lambda_j}\right)^{k_j}\right)} \]

Poisson

For positive integer data. For a class \(k\), the parameter is the mean and variance (\(\lambda_k\)). The density mass function is defined by:

\[ f_k(x) = \frac{\lambda^k}{k!}\exp{(-\lambda)} \]

NegativeBinomial

For positive integer data. For a class \(k\), parameters are the number of success (\(n_k\)) and the probability of success (\(p_k\)). The density mass function is defined by:

\[ f_k(x) = \frac{\Gamma(x+n_k)}{x! \Gamma(n_k)} p_k^{n_k}(1-p_k)^x \]

Multinomial

For categorical data. For a class \(k\), the model has \(M\) parameters \(p_{k,j},\, j=1,...,M\), where \(M\) is the number of modalities, corresponding to the probabilities to belong to the modality \(j\). \(p_{k,j},\, j=1,...,M\) must verify \(\sum_{j=1}^M p_{k,j} = 1\).

The density mass function is defined by:

\[ f_k(x = j) = \prod_{j=1}^K p_{k,j}^{a_j} \quad \text{with} \quad a_j = \begin{cases} 1 &\text{if } x = j \\ 0 &\text{otherwise} \end{cases} \]

The hyperparameter \(M\) does not require to be specified, it can be guess from the data. If tou want to specify it, add "nModality: M" in the appropriate field of the model object.

Rank_ISR

For ranking data. For a class \(k\), the two parameters are the central rank (\(\mu_k\)) and the probability of making a wrong comparison (\(\pi_k\)). See the article for more details. Ranks have their size \(M\) as hyperparameter. But it does not require to be specified, it can be guess from the data. If tou want to specify it, add "nModality: M" in the appropriate field of the description object.

Func_CS and Func_SharedAlpha_CS

For functional data. Between individuals, functional data can have different length and different time values. The model segments every functional and clusters them. The segmentation is performed using polynomial regressions on subpart of functionals. The model requires to indicate the desired number of subregressions \(S\), and the number of coefficients \(C\) used in each subregression, this number corresponds to the polynomial’s degree minus 1.

These hyperparameters must be specified by "nSub: S, nCoeff: C" in the appropriate field of the descriptor object. See the article for more details.

For a class \(k\) and a subregression \(s\), parameters are \(\alpha_{k,s,0}\) and \(\alpha_{k,s,1}\) the estimated coefficients of a logistic regression controlling the transition between subregressions, \(\beta_{k,s,1},...,\beta_{k,s,C}\) the estimated coefficient of the polynomial regression and \(\sigma_{k,s}\) the standard deviation of the residuals of the regression.

Func_SharedAlpha_CS is a variant of the Func_CS model with the alpha parameter shared between clusters. It means that the start and end of each subregression will be the same across the clusters.

Data Format

Real Data: Gaussian

Real values are saved with the dot as decimal separator. Missing data are indicated by a ?. Partial data can be provided through intervals denoted by [a:b] where a (resp. b) is a real or -inf (resp. +inf).

Real Positive Data: Weibull

Weibull data are real positive values with the dot as decimal separator. Missing data are indicated by a ?. Partial data can be provided through intervals denoted by [a:b] where a and b are positive reals (b can be +inf).

Counting Data: Poisson & NegativeBinomial

Counting data are positive integer. Missing data are indicated by a ?. Partial data can be provided through intervals denoted by [a:b] where a and b are positive integers (b can be +inf).

Categorical Data: Multinomial

Modalities must be consecutive integers with 1 as minimal value. Missing data are indicated by a ?. For partial data, a list of possible values can be provided by {a_1,...,a_j}, where a_i denotes a modality.

Categorical data before formatting:

varCateg1 varCateg2
married large
single small
status unknown medium
divorced small or medium
divorced or single large

after formatting:

Rank Data

The format of a rank is: o_1,..., o_j where o_1 is an integer corresponding to the the number of the object ranked in 1st position. For example: 4,2,1,3 means that the fourth object is ranked first then the second object is in second position and so on. Missing data can be specified by replacing and object by a ? or a list of potential object, for example: 4, {2 3}, {2 1}, ? means that the object ranked in second position is either the object number 2 or the object number 3, then the object ranked in third position is either the object 2 or 1 and the last one can be anything. A totally missing rank is spedified by a sequence of ? separated by commas, e.g. ?,?,?,? for a totally missing rank of length 4.

Functional Data: Func_CS & Func_SharedAlpha_CS

The format of a functional data is: time_1:value_1,..., time_j:value_j. Between individuals, functional data can have different length and different time values. In the case of a functional model, nSub: S, nCoeff: C must be indicated in the third row of the descriptor file. S is the number of subregressions in a functional data and C the number of coefficients of each regression (2 = linear, 3 = quadratic, …). Totally missing data are not supported. Time points with missing values must not be included.

For example if you have a time vector 1 2 3 4 5 6 8 12 and a value vector 0.8 0.6 0.88 0.42 0.62 0.75 0.72 0.66 for one individual, the MixtComp format is 1:0.8,2:0.6,3:0.88,4:0.42,5:0.62,6:0.75,8:0.72,12:0.66.

Missing Data Summary

Multinomial Gaussian Poisson NegativeBinomial Weibull Rank_ISR Func_CS LatentClass
Completely missing ? ? ? ? ? ?,?,?,? ?
Finite number of values {a,b,c} 4,{1 2},3,{1 2} {a,b,c}
Bounded interval [a:b] [a:b] [a:b] [a:b]
Right bounded interval [-inf:b]
Left bounded interval [a:+inf] [a:+inf] [a:+inf] [a:+inf]

(Semi-)Supervised Clustering

To perform a (semi-)supervised clustering, user can add a variable named z_class (with eventually some missing values) with "LatentClass" as model. Missing data are indicated by a ?. For partial data, a list of possible values can be provided by {a_1,...,a_j}, where a_i denotes a class number.