Add data.frame -> shinyTree helper function

[trestletech]

From #4

It would also be nice to add a small helper function that creates a nested list object, based on a data-frame which holds the hierarchical structure of the data to be visualized by jsTree. In the end I used a bunch of lapply calls to construct a nested list object which was a bit cumbersome. I guess better alternatives exist...

More details in comments to the original issue.

[JasperSch]

I came up with this function inspired by this discussion https://community.rstudio.com/t/how-to-recursively-apply-a-function-to-a-nested-list-of-data-frames/70324/3. I can make a PR if it looks good.

Function + MWE:

library(shiny)
library(shinyTree)

# Recursively apply function to all data frames in a nested list
dfrapply <- function(object, f, ...) {
  if (inherits(object, "data.frame")) {
    return(f(object, ...))
  }
  if (inherits(object, "list")) {
    return(lapply(object, function(x) dfrapply(x, f, ...)))
  }
  stop("List element must be either a data frame or another list")
}


#' convert a data.frame to nested list representing a tree
#' @param df data.frame or data.table
#' @param hierarchy character ordered vector of column names defining the hierarchy
#' @return nested list
dfToTree <- function(df, hierarchy){ 
  l <- df
  for(c in hierarchy){
    l <- dfrapply(
        object = l,
        f = function(x){
          split(x, x[[c]], drop = TRUE)
        }
    )
  }
  dfrapply(l, function(x){""})
}


ui <- fluidPage(shinyTree("myTree"))
server <- function(input,output,session){
  df <- data.frame(Titanic)
  tree <- dfToTree(df, c("Sex", "Class", "Survived"))
  output$myTree <- renderTree({tree})
}
shinyApp(ui,server)

[bellma-lilly]

@JasperSch This is very nice. I would lean toward this being an example rather than a member of the package because I think often people might want to customize it (to include the column name for example).

On the other hand, if we had a dfToTree and a treeToDF, maybe it makes sense to have those as member functions.

I would like to hear what other people think, and then I hope you will move forward with a PR.

[JasperSch]

@bellma-lilly Thank you for the feedback. Both options seem reasonable to me. I'll be glad to contribute once a decision is made.

[JasperSch]

Regarding treeToDf, this is what I'm currently using:

# @param tree nested list
# @param hierarchy ordered character vector with column name of each level
# @return data.frame
treeToDf <- function(tree, hierarchy){
  for(column in hierarchy){
    col <- sapply(tree, names)
    col <- lapply(col, function(x){
          if(is.null(x)){
            NA
          } 
          else {
            x
          }          
        })
    col <- unlist(col)
    df[,column] <- col 
    tree <- unlist(tree, recursive = FALSE) 
  }
df
}

# in shiny:
tree <- get_selected(input$tree, format = c("slices"))
df <- treeToDf(tree, hierarchy)

[bellma-lilly]

@JasperSch It looks like we aren't going to get a lot of feedback. I'll leave the decision to you, as it is your code. I look forward to adding it to the package. Thanks for your contribution!

[trafficonese]

Great work @JasperSch! I think the function definitely deserves to be included in the package as I think it would be easier to transform a dataframe into the structure the function needs than to build the list itself.

So, I think it would definitely benefit most of the users, especially newcomers. If you want to create your custom list, you can still do so.

[dommens]

Great work! Thanks a lot.