Movatterモバイル変換

Simplivariate Component Analysis monitoring function factory for GA progress

Description

Creates a monitoring function that prints the current generation and the best fitness scoreto the console at specified intervals. Intended for use as amonitor function in GA runs.

Usage

SCAMonitorFactory(interval = 100)

Arguments

Value

A monitoring function that can be used with GA. The returned function takes a GA objectand prints progress information at the specified interval.

Examples

# Create monitor that prints every 100 generations (default)monitor <- SCAMonitorFactory()# ga(..., monitor = monitor)# Create monitor that prints every 50 generationsmonitor <- SCAMonitorFactory(50)# ga(..., monitor = monitor)

Generate an additive approximation of a data matrix

Description

Constructs a matrix with an additive structure based on the row means,column means, and the grand mean of the original matrix.

Usage

additiveMatrix(mat)

Arguments

Details

The result is an approximationA of the original matrix, where:

A[i,j] = rowMean[i] + colMean[j] - grandMean

This model captures main additive effects of rows and columns, commonlyused in exploratory data analysis or baseline modeling.

Value

A numeric matrix of the same dimension asmat, approximating itusing an additive model.

Examples

m <- matrix(c(1, 2, 3, 4), nrow = 2)additiveMatrix(m)

Additive Pattern Fitter

Description

Fits an additive pattern to matrix data using training data specified by a mask.The pattern is based on row and column main effects, creating an additive modelwhere each cell value is the sum of a row effect and column effect minus the overall mean.

Usage

additiveMatrixFitter(mat, trainMask)

Arguments

Details

The function creates an additive pattern by:

• Masking non-training cells as NA to compute statistics only on training data

• Computing overall mean of training data

• Computing row means and column means from training data

• Replacing non-finite means with overall mean as fallback

• Creating additive pattern as outer sum of row and column effects minus overall mean

Value

Numeric matrix of same dimensions as mat containing the fitted additive pattern

Cross-Validation of Simplivariate Component Patterns

Description

Performs pure cross-validation over specified patterns with mandatory fitters.This function evaluates different pattern fitting models using cross-validationto determine the best model for a given data subset.

Usage

componentCVPatterns(  df,  rows,  cols,  patternFunctions,  patternFitters,  preferenceOrder = names(patternFunctions),  nRepeats = 40,  testFraction = 0.2,  minCellsForModels = 25,  parsimonyMargin = 0.05,  requireFitters = TRUE,  verbose = FALSE)

Arguments

Value

A list containing:

Default Pattern Fitters for SIMPLICA

Description

Returns a list of default pattern fitting functions used in SIMPLICA.These fitters estimate different types of patterns (constant, additive, multiplicative)from matrix data using specified training cells.

Usage

defaultPatternFitters()

Details

Each fitter function takes two arguments:

• mat: Numeric matrix containing the data to fit

• trainMask: Logical matrix indicating which cells to use for training

All fitters return a fitted matrix of the same dimensions as the input.

Value

A named list containing pattern fitting functions:

• constant: Fits a constant value (mean of training data)

• additive: Fits an additive pattern using additiveMatrixFitter

• multiplicative: Fits a multiplicative pattern using multiplicativeMatrixFitter

Examples

# Retrieve default pattern fittersfitters <- defaultPatternFitters()# Add a custom diagonal pattern fitterdiagonalFitter <- function(mat, trainMask) {  # Extract diagonal values from training data  minDim <- min(nrow(mat), ncol(mat))  diagIndices <- cbind(1:minDim, 1:minDim)  # Only use diagonal elements that are in the training mask  validDiag <- trainMask[diagIndices]  if (any(validDiag)) {    diagVal <- mean(mat[diagIndices][validDiag])  } else {    diagVal <- mean(mat[trainMask])  # fallback to overall mean  }  matrix(diagVal, nrow = nrow(mat), ncol = ncol(mat))}# Extend the list with your own patternfitters$diagonal <- diagonalFitter

Default pattern generators for SIMPLICA

Description

Returns a named list of default matrix approximation functions used to score component patterns.Each function must take a numeric matrix (i.e., a component: subset of rows and columns)and return a matrix of the same dimensions that approximates the original matrixaccording to a specific structural pattern (e.g., constant, additive, etc.).

Usage

defaultPatternFunctions()

Details

This list can be passed tofitness2() via thepatternFunctions argument.Users can extend or override the default patterns by modifying the returned list.

Requirements for pattern functions

Custom pattern functions must:

• Take a numeric matrix as input.

• Return a numeric matrix of the same dimensions.

• Be compatible withsum(abs(...)) andsum((...)^2) operations for fitness scoring.

Value

A named list of functions, each representing a matrix approximation method.

Examples

# Retrieve default pattern functionspatterns <- defaultPatternFunctions()# Add a custom pattern based on diagonal structurediagonalPattern <- function(m) {  diagVal <- mean(diag(as.matrix(m)))  matrix(diagVal, nrow = nrow(m), ncol = ncol(m))}# Extend the list with your own patternpatterns$diagonal <- diagonalPattern

Fitness function with automatic pattern selection per Simplivariate Component

Description

Fitness function with automatic pattern selection per Simplivariate Component

Usage

fitness(  string,  df,  dfMean,  penalty,  patternFunctions = defaultPatternFunctions(),  returnPatterns = FALSE,  ...)

Arguments

Value

Either total fitness (numeric), or list(fitness, componentPatterns) if returnPatterns = TRUE.

Compute best pattern-based fitness for a single Simplivariate Component

Description

Compute best pattern-based fitness for a single Simplivariate Component

Usage

fitnessForOneComponent(mat, dfMean, patternFunctions, penalty)

Arguments

Value

Numeric fitness value (higher is better)

Examples

m <- matrix(rnorm(100, mean = 10), nrow = 10)f <- fitnessForOneComponent(m, mean(m), defaultPatternFunctions(),               c(constant = 0, additive = 1.0, multiplicative = 0))

Integer mutation operator for genetic algorithms

Description

Applies mutation to a selected parent vector by replacing each gene with a random value(within bounds) with a given mutation probability. Used in integer-encoded GAs.

Usage

gaintegerMutation(object, parent, ...)

Arguments

Value

A numeric vector representing the mutated individual.

One-point crossover operator for integer-encoded genetic algorithms

Description

Performs one-point crossover on two parent individuals. A single crossover point is selected,and all genes before (and including) that point are exchanged between the parents.

Usage

gaintegerOnePointCrossover(object, parents, ...)

Arguments

Value

A list with two elements:

children

A 2-row matrix of the resulting offspring.

fitness

A numeric vector ofNA values to be replaced by fitness evaluation.

GA Integer Population Factory

Description

Creates a factory function for generating initial populations for genetic algorithmswith integer chromosomes, where a specified fraction of variables are set to zero.

Usage

gaintegerPopulationFactory(zeroFraction, verbose = FALSE)

Arguments

Value

A function that takes a GA object and returns an initial population matrix

Global Variables Declaration

Description

This file declares global variables used in the SIMPLICA package to avoidR CMD check notes about "no visible binding for global variable".These variables are typically used within ggplot2 aesthetics and data manipulationfunctions where they refer to column names in data frames created during execution.

Generate an multiplicative approximation of a data matrix

Description

Approximates a data matrix using a low-rank multiplicative model based ona fixed outer product of centered row and column effects.

Usage

multiplicativeMatrix(mat)

Arguments

Details

The model assumes:

M[i,j] = mu + a * rowEffect[i] * colEffect[j]

wheremu is the overall mean of the input matrix, androwEffect andcolEffect are centered integer sequences.

Value

A numeric matrix of the same size asmat, containing the fitted values.

Multiplicative Pattern Fitter

Description

Fits a multiplicative pattern to matrix data using training data specified by a mask.The pattern is based on the outer product of centered row and column effects, creatinga bilinear surface that can capture multiplicative interactions between rows and columns.

Usage

multiplicativeMatrixFitter(mat, trainMask)

Arguments

Details

The function creates a multiplicative pattern by:

• Computing the mean of training data as baseline

• Creating centered row effects (0 to nRows-1, mean-centered)

• Creating centered column effects (0 to nCols-1, mean-centered)

• Taking outer product to form bilinear pattern

• Fitting scaling coefficient using least squares on training data

• Returning baseline plus scaled pattern

Value

Numeric matrix of same dimensions as mat containing the fitted multiplicative pattern

Plot non-contiguous simplivariate components by pattern type, with optional reordering

Description

Visualizes GA-detected simplivariate components on the original matrix as outlined cells,colored by pattern type.

Usage

plotComponentResult(  df,  string,  componentPatterns,  componentScores,  scoreCutoff = 0,  showAxisLabels = TRUE,  showComponentLabels = TRUE,  title = "Detected Components",  rearrange = FALSE,  grayscale = TRUE)

Arguments

Value

ggplot object

Print method for summaryComponents

Description

Print method for summaryComponents

Usage

## S3 method for class 'summaryComponents'print(x, showDetails = FALSE, maxLinesDetails = 200L, ...)

Arguments

Value

The input objectx, invisibly. Called for its side effect ofprinting a formatted component summary to the console.

SIMPLICA: Simultaneous Identification of Simplivariate Components

Description

Implements the SIMPLICA algorithm to identify Simplivariate Componentsin data matrices using a genetic algorithm. These components are related toclusters or biclusters, but defined here in terms of specific structuralpatterns (constant, additive, multiplicative, or user-defined).

Usage

simplica(  df,  maxIter = 2000,  popSize = 300,  pCrossover = 0.6,  pMutation = 0.03,  zeroFraction = 0.9,  elitism = 100,  numSimComp = 5,  verbose = FALSE,  mySeeds = 1:5,  interval = 100,  penalty = c(constant = 0, additive = 1, multiplicative = 0),  patternFunctions = defaultPatternFunctions(),  doSimplicaCV = TRUE,  cvControl = NULL)

Arguments

Value

A list with:

• best: simplica object (includes original GA result; if doSimplicaCV=TRUE, also componentPatternsUpdated and componentAudit)

• raw: list of"ga" objects (one per seed, from the GA package)

References

Hageman, J. A., Wehrens, R., & Buydens, L. M. C. (2008). "Simplivariate Models: Ideas and First Examples." PLoS ONE, 3(9), e3259.doi:10.1371/journal.pone.0003259

Madeira, S. C., & Oliveira, A. L. (2004). "Biclustering Algorithms for Biological Data Analysis: A Survey." IEEE/ACM Transactions on Computational Biology and Bioinformatics, 1(1), 24–45.doi:10.1109/TCBB.2004.2

Examples

data("simplicaToy")# Minimal run just to demonstrate function usage, run with default GA parametersfit <- simplica(df = simplicaToy$data,                 maxIter = 200,                popSize = 50,                mySeeds = 1,                elitism = 1,                verbose = TRUE)plotComponentResult(df = simplicaToy$data,                    string            = fit$best$string,                    componentPatterns = fit$best$componentPatternsUpdated,                    componentScores   = fit$best$componentScores,                    showAxisLabels    = FALSE,                    title             = "SIMPLICA on simplicaToy",                    scoreCutoff       = 25000)

Test Simplivariate Components with Cross-Validation Pattern Selection

Description

This function performs cross-validation-based pattern testing for Simplivariate Componentsin a SIMPLICA object. It evaluates different pattern functions using cross-validationand selects the best performing pattern for each component. Fitters are requiredfor all patterns with no fallback options.

Usage

simplicaCV(  foundObject,  df,  patternFunctions = defaultPatternFunctions(),  patternFitters = defaultPatternFitters(),  preferenceOrder = names(patternFunctions),  nRepeats = 40,  testFraction = 0.2,  minCellsForModels = 25,  parsimonyMargin = 0.05,  requireFitters = TRUE,  updateObject = TRUE,  verbose = FALSE,  ignoreNaComponents = TRUE)

Arguments

Details

The function performs the following steps:

• Validates the input simplica object and data dimensions

• Checks that all pattern functions have corresponding fitters

• For each simplivariate component, performs cross-validation pattern evaluation

• Selects the best performing pattern based on RMSE and parsimony

• Updates component patterns and provides detailed test information

Value

IfupdateObject = TRUE, returns the inputsimplica objectwith two new fields:

componentPatternsUpdated

Character vector with the selectedpattern per component after cross-validation. If a component is skipped or empty,the entry isNA.

componentAudit

Data frame containing detailed cross-validationresults for each component, with the following columns:

componentId

Numeric ID of the component.

originalPattern

Pattern label originally assigned.

selectedPattern

Pattern chosen after CV-based evaluation.

reason

Explanation of why a pattern was selected or skipped.

nRows,nCols,nCells

Dimensions of the component.

nRepeats,testFraction,parsimonyMargin

CV settings used.

cvMean_<pattern>

Mean RMSE over CV folds for each tested pattern.

cvSd_<pattern>

Standard deviation of RMSE across CV folds.

winFrac_<pattern>

Fraction of CV repeats where the pattern was the best performer.

IfupdateObject = FALSE, returns a list with the same two elements(componentPatternsUpdated,componentAudit).

Toy matrix with one multiplicative and one additive bicluster

Description

A small 30×60 matrix to demonstrate SIMPLICA in a controlled setting.Contains one multiplicative and one additive simplivariate component (non-overlapping).

Usage

data(simplicaToy)

Format

A list with three elements:

data

numeric matrix of dimension30 \times 60

trueComponents

list of length 2 withtype,rows,cols

description

character string

Examples

data("simplicaToy")str(simplicaToy)image(t(simplicaToy$data))

Summarize GA-found Simplivariate Components

Description

Compute a tidy summary of simplivariate Components found by SIMPLICA.Returns a data.frame with class"summaryComponents" and an attribute holdingrow/column index lists for printing details.

Usage

summaryComponents(results, scoreCutoff = 0)

Arguments

Value

A data.frame with columns:componentId,pattern,score,rows,cols,size;class isc("summaryComponents","data.frame"). The attribute"indices" stores a list with per-componentrowIdx andcolIdx.