R/plot_neighbors.r

In codymarquart/LSAfun: Applied Latent Semantic Analysis (LSA) Functions

################################################
##### Plot neighbors ##########################

#' @export plot_neighbors
#' 
plot_neighbors <- function(x,n,connect.lines=0,
                            start.lines=T,method="PCA",dims=3,
                            axes=F,box=F,cex=1,
                            alpha=.5, col="black", 
                            tvectors=tvectors,breakdown=TRUE,
                            ...) {
  ### Compute neighbors
  if(!(dims %in% 2:3)){
    stop("Please set dim to 2 or 3")
  }
  
  if(class(tvectors) == "matrix"){
    if(class(x) == "character"){
      if(breakdown==TRUE){satz1 <- breakdown(x)}  
      if(breakdown==FALSE){satz1 <- x}  
        
      satz1split <- strsplit(satz1,split=" ")[[1]]
      used1 <- satz1split[satz1split %in% rownames(tvectors)]
        
      if(length(used1) > 1){satz1vec <- colSums(tvectors[used1,])}
      if(length(used1) == 1){satz1vec <- (tvectors[used1,])}
        
      if(length(used1)==0){return(warning(
        "no element of x found in rownames(tvectors)"))
      }
        
      near <- names(neighbors(satz1vec,n, tvectors=tvectors, breakdown=breakdown))
    }
      
    if(class(x) == "numeric") {
      satz1vec  <- x
      near      <- names(neighbors(x,n,tvectors=tvectors, breakdown=breakdown))
    }

    nearwords <- paste(near,collapse=" ")
    cos.near  <- multicos(nearwords,tvectors=tvectors, breakdown=FALSE)
    
    #### Add Phrase to diagram
    if((class(x) == "character")){
      if(length(satz1split) > 1){
        letters     <- unlist(strsplit(x,""))[1:15]
        letters     <- letters[!is.na(letters)]
        expressionW <- paste(letters,sep="",collapse="")
        if(expressionW != x){
          expressionW <- paste(expressionW,"[...]",sep="")
        }
        
        rownames(cos.near)[1] <- expressionW
        colnames(cos.near)[1] <- expressionW
      }
    }
    
    if(class(x) == "numeric"){
      rownames(cos.near)[1] <- "Input Vector"
      colnames(cos.near)[1] <- "Input Vector"
    }
    
    ## Reduce dimensions
    if(method=="PCA"){
      pca1 <- princomp(covmat=cos.near)
      L    <- loadings(pca1) %*% diag(pca1$sdev)
      Lt   <- varimax(L[,1:dims])$loadings
    }
    
    if(method=="MDS"){
      dissim <- 1 - cos.near
      mds1 <- cmdscale(dissim,eig=TRUE, k=dims) 
      Lt   <- mds1$points   
    }
    
    Lt <- as.data.frame(Lt[,])
    
    if(dims==2){colnames(Lt) <- c("x","y")}
    if(dims==3){colnames(Lt) <- c("x","y","z")}
    
    Lt$words     <- rownames(Lt)
    Lt$words2     <- iconv(Lt$words, to="ASCII//TRANSLIT")

    ## Plot 2d
    if(dims==2){
      plot(Lt$x,Lt$y,xlab="Dimension 1",ylab="Dimension 2",pch=20,type="n",
         xlim=c(min(Lt$x)-0.1,max(Lt$x)+0.1),ylim=c(min(Lt$y)-0.1,max(Lt$y)+0.1))
      with(Lt,points(x,y,cex=.6,pch=20))
      with(Lt,text(x,y,words2,cex=cex))
    }
    
    ## Plot 3d
    if(dims == 3){
      if (!requireNamespace("rgl", quietly = TRUE) || !exists('plot3d') || !exists('text3d') || !exists('segments3d')) {
        return(warning("`rgl` needed for this function to work with 3 dimensions. Please install/load it."))
      }

      with(Lt,plot3d(x,y,z,box=box,axes=axes,xlab="",ylab="",zlab="",xakt="n",yakt="n",zakt="n",col="black",...))
      with(Lt,text3d(x,y,z,words2))
    
      ### Rainbow palette
      palette <- rainbow(101,start=0,end=0.95)
    
      ### connect.lines : X nearest
      ### connect.lines use argument [2] from alpha and col!
    
      if(connect.lines == "all"){
        #### Create all pairwise combinations as vector
        comb  <- combn(rownames(Lt),m=2)
        combs <- c(combn(rownames(Lt),m=2))
      
        #### Pairwise combination similarities
        pwsim  <- cos.near[lower.tri(cos.near)]
        pwsim2 <- rep(pwsim,each=2)
      
        #### Prepare dataframe for segments: Connected pairs ordered in rows
        segm <- Lt[combs,]
      
        #### Add Pairwise similarities
        segm$pwsim <- pwsim2
      
        #### Add colours
        colour      <- palette[round(pos(pwsim2*100)+1)]
        segm$colour <- colour
      
        #### Plot lines
        if(col != "rainbow"){suppressWarnings(segments3d(segm,alpha=(segm$pwsim)^2))}
        if(col == "rainbow"){suppressWarnings(segments3d(segm,alpha=(segm$pwsim)^2,col=segm$colour))}
      }
        
      if(length(alpha)==1){alpha <- rep(alpha,2)}
      if(length(col)==1 && col != "rainbow"){col <- rep(col,2)}
      
      if(class(connect.lines) == "numeric" && connect.lines > 0){
        ### Find nearest to each word  
        which.indices       <- t(apply(cos.near, 1, order, decreasing = T)[ 1:(connect.lines+1), ])
        which.indices       <- as.data.frame(which.indices[,-1])
        which.indices[,3:4] <- 1:n
        which.indices       <- which.indices[,c(3,1,4,2)]
        indices       <- as.vector(t(which.indices))
        
        ### Prepare Segments
        segm <- Lt[indices,]
        
        ### Add pairwise similarities
        pwsim <- vector(length=nrow(segm)/2)
        
        for(i in seq(1,nrow(segm),2)){
          pwsim[ceiling(i/2)] <- cos.near[segm[i,"words"],segm[(i+1),"words"]]
        }
        
        pwsim2 <- rep(pwsim,each=2)
        segm$pwsim <- pwsim2
        
        #### Add colours
        colour      <- palette[round(pos(pwsim2*100)+1)]
        segm$colour <- colour
        
        ### Plot
        if(col[1] != "rainbow" && alpha[1] != "shade") {
          suppressWarnings(segments3d(segm,alpha=alpha[2],col=col[2]))
        }
        if(col[1] != "rainbow" && alpha[1] == "shade") {
          suppressWarnings(segments3d(segm,alpha=(segm$pwsim)^2,col=col[2]))
        }
        if(col[1] == "rainbow") {
          suppressWarnings(segments3d(segm,alpha=(segm$pwsim)^2,col=segm$colour))
        }
      }  
        
      ### lines from original word to vectors
      ### start.lines use argument [1] from alpha and col!
      if(start.lines==T && connect.lines != "all"){
        steps                 <- vector(length=(2*n))
        steps[seq(1,2*n-1,2)] <- rep(1,n)
        steps[seq(2,2*n,2)]   <- 1:(n) 
        
        #### Prepare Segments
        segm0 <- Lt[steps,]
        pwsim0      <- rep(cos.near[1,],each=2)
        segm0$pwsim <- pwsim0
        colour0      <- palette[round(pos(pwsim0*100)+1)]
        segm0$colour  <- colour0
        
        if(col[1] != "rainbow" && alpha[1] != "shade"){
          suppressWarnings(segments3d(segm0,alpha=alpha[1],col=col[1]))
        }
        
        if(col[1] != "rainbow" && alpha[1] == "shade"){
          suppressWarnings(segments3d(segm0,alpha=(segm0$pwsim)^2,col=col[1]))
        }  
        
        if(col[1] == "rainbow"){
          suppressWarnings(segments3d(segm0,alpha=(segm0$pwsim)^2,col=segm0$colour))
        }
      }
    }
      
    Lt[,-which(colnames(Lt) %in% c("words","words2"))]
  } else {
    warning("tvectors must be a matrix!")
  }
}