R/mmpc.model.R

In MXM: Feature Selection (Including Multiple Solutions) and Bayesian Networks

# target: the target value
# sesObject: the outcome of the ses
# nisgnat: Number of signatures and generated models. It could be numeric from 1 to total number of signatures or "all" for all the 
## signatures. Default is 1.
mmpc.model <- function(target, dataset, wei = NULL, mmpcObject, test = NULL) {
  
  signature <- mmpcObject@selectedVars  
  if ( sum( is.na(signature) ) > 0 | length(signature) == 0 ) {
    mod <- paste("No associations were found, hence no model is produced.")
    signature <- NULL
    res <- list(mod = mod, signature = signature)  
    
  } else {
  
  if ( any(is.na(dataset) ) ) {
    warning("The dataset contains missing values (NA) and they were replaced automatically by the variable (column) median (for numeric) or by the most frequent level (mode) if the variable is factor")
    if ( is.matrix(dataset) )  {
      dataset <- apply( dataset, 2, function(x){ x[which(is.na(x))] = median(x, na.rm = TRUE) ; return(x) } ) 
    } else {
      poia <- which( is.na(dataset), arr.ind = TRUE )[2]
      for ( i in poia )  {
        xi <- dataset[, i]
        if ( is.numeric(xi) ) {                    
          xi[ which( is.na(xi) ) ] <- median(xi, na.rm = TRUE) 
        } else if ( is.factor( xi ) ) {
          xi[ which( is.na(xi) ) ] <- levels(xi)[ which.max( as.vector( table(xi) ) )]
        }
        dataset[, i] <- xi
      }
    }
  }
  
  if ( is.null(test) ) {  
    ci_test <- mmpcObject@test
  } else ci_test <- test 
  
    p <- length(signature)

    if ( ci_test == "testIndFisher" || ci_test == "testIndReg" ) {
      mod <- lm( target ~ ., data = as.data.frame(dataset[, signature ]), weights = wei )
      bic <- BIC(mod)       
    
    } else if ( ci_test == "tesIndMMReg" ) {
      mod <- MASS::rlm(target ~., data = data.frame(dataset[, signature ]), maxit = 2000, weights = wei )
      bic <- BIC( mod )       
	
    } else if ( ci_test == "testIndSpearman" || ci_test == "testIndRQ" ) {

      mod <- quantreg::rq( target ~., data = as.data.frame(dataset[, signature ]), weights = wei  )
  	  la <- logLik(mod)
      bic <-  - 2 * as.numeric( la ) + attr(la, "df") * log( length(target) )   

    } else if ( ci_test == "testIndBeta" ) {
      mod <- beta.mod( target, dataset[, signature ], wei = wei )
      bic <-  - 2 * mod$loglik + ( length( mod$be ) + 1 ) * log( length(target) )

    } else if ( ci_test == "testIndPois") {
      mod <- glm( target ~ ., data = data.frame(dataset[, signature ]), family = poisson, weights = wei )
      bic <- BIC( mod )
      
    } else if ( ci_test == "testIndQPois") {
      mod <- glm( target ~ ., data = data.frame(dataset[, signature ]), family = quasipoisson, weights = wei )
      bic <- NA

    } else if ( ci_test == "testIndQBinom") {
      mod <- glm( target ~ ., data = data.frame(dataset[, signature ]), family = quasibinomial, weights = wei )
      bic <- NA
      
    } else if ( ci_test == "testIndNB" ) {
      mod <- MASS::glm.nb( target ~ ., data = data.frame(dataset[, signature ]), weights = wei )
      bic <- BIC(mod)
      
    } else if ( ci_test == "testIndZIP" ) {
      mod <- zip.mod( target, dataset[, signature], wei = wei )
      bic <-  -2 * mod$loglik + ( length( coef(mod$be) ) + 1) * log( length(target) )
      
    } else if (ci_test == "testIndIGreg") {
      mod <- glm(target ~., data = data.frame( dataset[, signature] ), family = inverse.gaussian(log), weights = wei)
      bic <- BIC(mod)
      
    } else if ( is.matrix(target) & ci_test == "testIndMVreg" ) {
      if ( all(target > 0 & target < 1)  &  Rfast::Var( Rfast::rowsums(target) ) == 0 )   target = log( target[, -1]/(target[, 1]) ) 
      mod <- lm( target ~., data = data.frame(dataset[, signature ]), weights = wei )
      bic <- NULL
      
    } else if ( is.matrix(target)   &  ci_test == "testIndBinom" ) {
      mod <- glm( target[, 1] /target[, 2] ~., data = data.frame(dataset[, signature ]), weights = target[, 2], family = binomial )
      bic <- BIC(mod)
      
    } else if ( ci_test == "testIndGamma" ) {
      mod <- glm( target ~ ., data = data.frame(dataset[, signature ]), weights = wei, family = Gamma(link = log) )
      bic <- BIC(mod)
      
    } else if ( ci_test == "testIndNormLog" ) {
      mod <- glm( target ~ ., data = data.frame(dataset[, signature ]), weights = wei, family = gaussian(link = log) )
      bic <- BIC(mod)
      
    } else if ( ci_test == "testIndTobit" ) {
      mod <- survival::survreg( target ~ ., data = data.frame(dataset[, signature ]), weights = wei, dist = "gaussian" )
      bic <-  - 2 * as.numeric( logLik(mod) ) + ( length( mod$coefficients ) + 1 ) * log( NROW(dataset) )
      
    } else if (ci_test == "censIndCR") {
      mod <- survival::coxph( target ~ ., data = data.frame(dataset[, signature ]), weights = wei )
      bic <- BIC(mod)
      
    } else if (ci_test == "censIndWR") {
      mod <- survival::survreg( target ~ ., data = data.frame(dataset[, signature ]), weights = wei )
      bic <-  - 2 * as.numeric( logLik(mod) ) + ( length( mod$coefficients ) + 1 ) * log( NROW(dataset) )
      
    } else if (ci_test == "censIndER") {
      mod <- survival::survreg( target ~ ., data = data.frame(dataset[, signature ]), weights = wei, dist = "exponential" )
      bic <-  - 2 * as.numeric( logLik(mod) ) + ( length( mod$coefficients ) + 1 ) * log( NROW(dataset) )
      
    } else if (ci_test == "censIndLLR") {
      mod <- survival::survreg( target ~ ., data = data.frame(dataset[, signature ]), weights = wei, dist = "loglogistic" )
      bic <-  - 2 * as.numeric( logLik(mod) ) + ( length( mod$coefficients ) + 1 ) * log( NROW(dataset) )
      
    } else if (ci_test == "testIndClogit") {
      case <- as.logical(target[, 1]);  
      id <- target[, 2]
      mod <- survival::clogit(case ~ . + strata(id), data = data.frame( dataset[ , signature] ) )  ## wieghts are ignored here anyway
      bic <- BIC(mod)
      
    } else if ( ci_test == "testIndLogistic" ) {
      mod <- glm( target ~., data = as.data.frame(dataset[, signature ]), family = binomial, weights = wei ) 
      bic <- BIC(mod)

    } else if ( ci_test == "testIndMultinom" || ci_test == "gSquare" ) {
      mod <- nnet::multinom( target ~., data = as.data.frame(dataset[, signature ]), trace = FALSE, weights = wei )
      bic <- BIC(mod)
        
    } else if ( ci_test == "testIndOrdinal" ) {
      mod <- ordinal::clm( target ~., data = as.data.frame(dataset[, signature ]), weights = wei  )
      bic <- BIC(mod)    
    }
       
    if ( is.null( colnames(dataset) ) ) {
      names(signature) = paste("Var", signature, sep = " ")
    } else  names(signature) = colnames(dataset)[signature]
    
    signature <- c(signature, bic)
    names(signature)[length(signature)] = "bic"
    
    res <- list(mod = mod, signature = signature)  
    
  } ## end if ( sum( is.na(mmpcObject@selectedVars) ) > 0 ) 
  
  res
  
}