# Tour in loon An interactive tour in loon ```r l_tour( data, scaling = c("data", "variable", "observation", "sphere"), by = NULL, on, as.l_tour = TRUE, color = loon::l_getOption("color"), tour_path = tourr::grand_tour(), group = "color", start = NULL, slicing = FALSE, slicingDistance = NULL, numOfTours = 30L, interpolation = 40L, parent = NULL, envir = parent.frame(), ... ) ``` ## Arguments - `data`: a data frame with numerical data only - `scaling`: one of 'variable', 'data', 'observation', 'sphere', or 'none' to specify how the data is scaled. See Details - `by`: loon plot can be separated by some variables into multiple panels. This argument can take a `formula`, `n` dimensional state names (see `l_nDimStateNames`) an `n`-dimensional `vector` and `data.frame` or a `list` of same lengths `n` as input. - `on`: if the `x` or `by` is a formula, an optional data frame containing the variables in the `x` or `by`. If the variables are not found in data, they are taken from environment, typically the environment from which the function is called. - `as.l_tour`: return a `l_tour` object; see details - `color`: vector with line colors. Default is given by `l_getOption("color")`. - `tour_path`: tour path generator, defaults to 2d grand tour - `group`: only used for layers. As we scroll the bar, the layers are re-calculated. This argument is used to specify which state is used to set groups (i.e. "color", "linewidth", etc). - `start`: projection to start at, if not specified, uses default associated with tour path - `slicing`: whether to show a sliced scatter plot - `slicingDistance`: the slicing distance that if the distance between points and the projected plane is less than this distance, points will be preserved; else points will be invisible. The default is `NULL` and a suggested value will be given. See details - `numOfTours`: the number of tours - `interpolation`: the steps between two serial projections. The larger the value is, the smoother the transitions would be. - `parent`: a valid Tk parent widget path. When the parent widget is specified (i.e. not `NULL`) then the plot widget needs to be placed using some geometry manager like `tkpack` or `tkplace` in order to be displayed. See the examples below. - `envir`: the `environment` to use. - `...`: named arguments to modify the serialaxes states or layouts, see details. ## Returns an `l_tour` or an `l_tour_compound` object that one can query the `loon` states and a matrix projection vectors ## Details * tour_path is a tour generator; available tours are `grand_tour`, `dependence_tour`, `frozen_tour`, `guided_tour`, `planned_tour`, and etc * Argument `as.l_tour` * If set to `TRUE`, the function returns an `l_tour` (or an `l_tour_compound`) object. Essentially, this object is a list with the first element being a `loon` (Tcl) widget and the second element a matrix of projection vectors. The advantage of this setup is that the matrix of projection vectors can be easily accessed using the ```[``` function (or the `l_cget` function). However, a limitation is that it does not constitute a valid `loon` (Tcl) widget-calling `l_isLoonWidget` would return `FALSE`. Consequently, many of loon's functionalities remain inaccessible. * If set to `FALSE`, the function returns either a `loon` (Tcl) widget (where calling `l_isLoonWidget` would return `TRUE`) or an `l_compound` object. In this case, the matrix of projection vectors is not directly accessible from it. However, the `l_getProjection` function can be used to retrieve an estimated matrix of projection vectors. * The `scaling` state defines how the data is scaled. The axes display 0 at one end and 1 at the other. For the following explanation assume that the data is in a n x p dimensional matrix. The scaling options are then ||| |:--|:--| |variable|per column scaling| |observation|per row scaling| |data|whole matrix scaling| |sphere|transforming variables to principal components| * The default `slidingDistance` is suggested by Laa, U., Cook, D., & Valencia, G. (2020). First, find the maximum Euclidean distance of each observation (centralized), say `maxD`. Then, compute the "relative volume" that `vRel` = (`maxD`^(d - 2))/10, where `d` is the dimension of this data set. In the end, the suggested `slidingDistance` is given by `vRel`^(1/(d - 2)) ## Examples ```r if(interactive() && requireNamespace('tourr')) { # 2D projection fl <- tourr::flea[, 1:6] # different scaling will give very different projections # in this dataset, scaling 'variable' will give the best separation p <- l_tour(fl, scaling = 'variable', color = tourr::flea$species) l0 <- l_layer_hull(p, group = p["color"], color = "red", linewidth = 4) l1 <- l_layer_density2d(p) # a `l_tour` object class(p) # query the matrix of projection vectors proj <- p['projection'] # or `l_getProjection(p)` # suppose the scaling is still 'observation' new_xy <- as.matrix( loon::l_getScaledData(data = fl, scaling = 'observation')) %*% proj plot(new_xy, xlab = "V1", ylab = "V2", col = loon::hex12tohex6(p['color'])) # A higher dimension projection # turn the `tour` to 4 dimensional space s <- l_tour(fl, color = tourr::flea$species, scaling = "observation", tour_path = tourr::grand_tour(4L)) # set `as.l_tour` FALSE p <- l_tour(fl, scaling = 'observation', color = tourr::flea$species) class(p) ## ERROR ## p["projection"] # query the estimated matrix of projection vectors l_getProjection(p) ##### facet by region olive <- tourr::olive p <- with(olive, l_tour(olive[, -c(1, 2)], by = region, color = area)) } ``` ## See Also `l_getProjection`

l_tour function