R/linereg.R

In nitinsverige/casparnitinlinreg: This packages is the classical linear regression.

#' Title
#'
#' @field dep matrix.
#' @field indep matrix.
#' @field betahat matrix.
#' @field predicted matrix.
#' @field residuals matrix.
#' @field inv_sqr_dep matrix.
#' @field deg_fre numeric.
#' @field sigmasq numeric.
#' @field var_betas numeric.
#' @field t_values numeric.
#' @field p_values numeric.
#' @field data ANY.
#'
#' @return
#' @export linreg
#' @exportClass linreg
#'
#' @examples
linreg <- setRefClass("linreg",
                      fields = list(dep = "matrix", indep = "matrix", betahat = "matrix",
                                    predicted = "matrix", residuals = "matrix",inv_sqr_dep = "matrix",
                                    deg_fre = "numeric", sigmasq="numeric", var_betas = "numeric",
                                    t_values = "numeric",p_values = "numeric" , data = "ANY" ),
                      methods = list(
                        coef = function(){
                          "Method coef"

                          sqr_dep<-t(indep)%*%indep #creates the square matrix

                          inv_sqr_dep<<-solve(sqr_dep) # Creates an inverse matrix to the square matrix

                          t_indep_dep<-t(indep)%*%dep #creates a matrix that is solvebal

                          betahat <<- inv_sqr_dep%*%t_indep_dep # Creates betahat for the regression

                        },
                        pred = function(){
                          "Method for prediction"
                          predicted <<- indep%*%betahat
                        },
                        resid = function(){residuals<<- predicted - dep},
                        dfre = function(){ deg_fre<<- nrow(indep) - ncol(indep)  },
                        sigma= function(){   sig <- (t(residuals)%*%residuals)/deg_fre
                        sigmasq<<- as.numeric(sig)
                        },
                        vbetas=function(){  var <-sigmasq*inv_sqr_dep
                        var_betas <<- sqrt( abs(colSums(var)) ) # Sums the variance for each colum.
                        },
                        tvalue=function(){
                          "method t-value"
                          t_values <<- as.numeric( betahat/ var_betas )
                        p_values <<- pt(t_values, deg_fre)},
                        plot  = function(){
                          "plots the residuals"
                          df<-data.frame(predicted, residuals)
                        ggplot(data = df) + aes(x =predicted, y=residuals, group=2) + geom_point()+geom_abline(intercept = 0, slope = 0 ) },
                        model = function(y,x,...){
                          "Formula for reggression"

                          if(is.vector(y)& dim(y)[2]==1){ y <- as.matrix(y)} # recognize the variable and puts it to the right type.
                          if(is.matrix(x)){}

                        }

                      ) )