\name{E_interaction} \alias{panel.identify} \alias{panel.identify.qqmath} \alias{panel.identify.cloud} \alias{panel.link.splom} \alias{panel.brush.splom} \alias{trellis.focus} \alias{trellis.unfocus} \alias{trellis.switchFocus} \alias{trellis.panelArgs} \alias{trellis.vpname} \alias{trellis.grobname} \concept{interaction} \concept{augment} \title{Functions to Interact with Lattice Plots} \description{ The classic Trellis paradigm is to plot the whole object at once, without the possibility of interacting with it afterwards. However, by keeping track of the grid viewports where the panels and strips are drawn, it is possible to go back to them afterwards and enhance them one panel at a time. These functions provide convenient interfaces to help in this. Note that these are still experimental and the exact details may change in future. } \usage{ panel.identify(x, y = NULL, subscripts = seq_along(x), labels = subscripts, n = length(x), offset = 0.5, threshold = 18, ## in points, roughly 0.25 inches panel.args = trellis.panelArgs(), \dots) panel.identify.qqmath(x, distribution, groups, subscripts, labels, panel.args = trellis.panelArgs(), \dots) panel.identify.cloud(x, y, z, subscripts, perspective, distance, xlim, ylim, zlim, screen, R.mat, aspect, scales.3d, \dots, panel.3d.identify, n = length(subscripts), offset = 0.5, threshold = 18, labels = subscripts, panel.args = trellis.panelArgs()) panel.link.splom(threshold = 18, verbose = getOption("verbose"), \dots) panel.brush.splom(threshold = 18, verbose = getOption("verbose"), \dots) trellis.vpname(name = c("position", "split", "split.location", "toplevel", "figure", "panel", "strip", "strip.left", "legend", "main", "sub", "xlab", "ylab", "page"), column, row, side = c("left", "top", "right", "bottom", "inside"), clip.off = FALSE, prefix) trellis.grobname(name, prefix) trellis.focus(name, column, row, side, clip.off, highlight = interactive(), \dots, guess = TRUE, verbose = getOption("verbose")) trellis.switchFocus(name, side, clip.off, highlight, \dots) trellis.unfocus() trellis.panelArgs(x, packet.number) } \arguments{ \item{x, y, z}{ variables defining the contents of the panel. In the case of \code{trellis.panelArgs}, a \code{"trellis"} object. } \item{n}{ the number of points to identify by default (overridden by a right click) } \item{subscripts}{ an optional vector of integer indices associated with each point. See details below. } \item{labels}{ an optional vector of labels associated with each point. Defaults to \code{subscripts} } \item{distribution, groups}{ typical panel arguments of \code{\link{panel.qqmath}}. These will usually be obtained from \code{panel.args} } \item{offset}{ the labels are printed either below, above, to the left or to the right of the identified point, depending on the relative location of the mouse click. The \code{offset} specifies (in "char" units) how far from the identified point the labels should be printed. } \item{threshold}{ threshold in grid's \code{"points"} units. Points further than these from the mouse click position are not considered } \item{panel.args}{ list that contains components names \code{x} (and usually \code{y}), to be used if \code{x} is missing. Typically, when called after \code{trellis.focus}, this would appropriately be the arguments passed to that panel. } \item{perspective, distance, xlim, ylim, zlim, screen, R.mat, aspect, scales.3d}{ arguments as passed to \code{\link{panel.cloud}}. These are required to recompute the relevant three-dimensional projections in \code{panel.identify.cloud}. } \item{panel.3d.identify}{ the function that is responsible for the actual interaction once the data rescaling and rotation computations have been done. By default, an internal function similar to \code{panel.identify} is used. } \item{name}{ character string indicating which viewport or grob we are looking for. Although these do not necessarily provide access to all viewports and grobs created by a lattice plot, they cover most that users might find interesting. \code{trellis.vpname} and \code{trellis.focus} deal with viewport names only, and only accept the values explicitly listed above. \code{trellis.grobname} is meant to create names for grobs, and can currently accept any value. If \code{name}, as well as \code{column} and \code{row} is missing in a call to \code{trellis.focus}, the user can click inside a panel (or an associated strip) to focus on that panel. Note however that this assumes equal width and height for each panel, and may not work when this is not true. When \code{name} is \code{"panel"}, \code{"strip"}, or \code{"strip.left"}, \code{column} and \code{row} must also be specified. When \code{name} is \code{"legend"}, \code{side} must also be specified. } \item{column, row}{ integers, indicating position of the panel or strip that should be assigned focus in the Trellis layout. Rows are usually calculated from the bottom up, unless the plot was created with \code{as.table=TRUE} } \item{guess}{ logical. If \code{TRUE}, and the display has only one panel, that panel will be automatically selected by a call to \code{trellis.focus}. } \item{side}{ character string, relevant only for legends (i.e., when \code{name="legend"}), indicating their position. Partial specification is allowed, as long as it is unambiguous. } \item{clip.off}{ logical, whether clipping should be off, relevant when \code{name} is \code{"panel"} or \code{"strip"}. This is necessary if axes are to be drawn outside the panel or strip. Note that setting \code{clip.off=FALSE} does not necessarily mean that clipping is on; that is determined by conditions in effect during printing. } \item{prefix}{ character string acting as a prefix, meant to distinguish otherwise equivalent viewports in different plots. This only becomes relevant when a particular page is occupied by more than one plot. Defaults to the value appropriate for the last \code{"trellis"} object printed, as determined by the \code{prefix} argument in \code{\link{print.trellis}}. Users should not usually need to supply a value for this argument (see note below), however, if supplied explicitly, this has to be a valid R symbol name (briefly, it must start with a letter or a period followed by a letter) and must not contain the grid path separator (currently \code{"::"}) } \item{highlight}{ logical, whether the viewport being assigned focus should be highlighted. For \code{trellis.focus}, the default is \code{TRUE} in interactive mode, and \code{trellis.switchFocus} by default preserves the setting currently active. } \item{packet.number}{ integer, which panel to get data from. See \code{\link{packet.number}} for details on how this is calculated } \item{verbose}{ whether details will be printed } \item{\dots}{ For \code{panel.identify.qqmath}, extra parameters are passed on to \code{panel.identify}. For \code{panel.identify}, extra arguments are treated as graphical parameters and are used for labelling. For \code{trellis.focus} and \code{trellis.switchFocus}, these are used (in combination with \code{\link{lattice.options}}) for highlighting the chosen viewport if so requested. Graphical parameters can be supplied for \code{panel.link.splom}. } } \details{ \code{panel.identify} is similar to \code{\link{identify}}. When called, it waits for the user to identify points (in the panel being drawn) via mouse clicks. Clicks other than left-clicks terminate the procedure. Although it is possible to call it as part of the panel function, it is more typical to use it to identify points after plotting the whole object, in which case a call to \code{trellis.focus} first is necessary. \code{panel.link.splom} is meant for use with \code{\link{splom}}, and requires a panel to be chosen using \code{trellis.focus} before it is called. Clicking on a point causes that and the corresponding proections in other pairwise scatter plots to be highlighted. \code{panel.brush.splom} is a (misnamed) alias for \code{panel.link.splom}, retained for back-compatibility. \code{panel.identify.qqmath} is a specialized wrapper meant for use with the display produced by \code{\link{qqmath}}. \code{panel.identify.qqmath} is a specialized wrapper meant for use with the display produced by \code{\link{cloud}}. It would be unusual to call them except in a context where default panel function arguments are available through \code{trellis.panelArgs} (see below). One way in which \code{panel.identify} etc. are different from \code{\link{identify}} is in how it uses the \code{subscripts} argument. In general, when one identifies points in a panel, one wants to identify the origin in the data frame used to produce the plot, and not within that particular panel. This information is available to the panel function, but only in certain situations. One way to ensure that \code{subscripts} is available is to specify \code{subscripts = TRUE} in the high level call such as \code{xyplot}. If \code{subscripts} is not explicitly specified in the call to \code{panel.identify}, but is available in \code{panel.args}, then those values will be used. Otherwise, they default to \code{seq_along(x)}. In either case, the final return value will be the subscripts that were marked. The process of printing (plotting) a Trellis object builds up a grid layout with named viewports which can then be accessed to modify the plot further. While full flexibility can only be obtained by using grid functions directly, a few lattice functions are available for the more common tasks. \code{trellis.focus} can be used to move to a particular panel or strip, identified by its position in the array of panels. It can also be used to focus on the viewport corresponding to one of the labels or a legend, though such usage would be less useful. The exact viewport is determined by the \code{name} along with the other arguments, not all of which are relevant for all names. Note that when more than one object is plotted on a page, \code{trellis.focus} will always go to the plot that was created last. For more flexibility, use grid functions directly (see note below). After a successful call to \code{trellis.focus}, the desired viewport (typically panel or strip area) will be made the \sQuote{current} viewport (plotting area), which can then be enhanced by calls to standard lattice panel functions as well as grid functions. It is quite common to have the layout of panels chosen when a \code{"trellis"} object is drawn, and not before then. Information on the layout (specifically, how many rows and columns, and which packet belongs in which position in this layout) is retained for the last \code{"trellis"} object plotted, and is available through \code{trellis.currentLayout}. \code{trellis.unfocus} unsets the focus, and makes the top level viewport the current viewport. \code{trellis.switchFocus} is a convenience function to switch from one viewport to another, while preserving the current \code{row} and \code{column}. Although the rows and columns only make sense for panels and strips, they would be preserved even when the user switches to some other viewport (where row/column is irrelevant) and then switches back. Once a panel or strip is in focus, \code{trellis.panelArgs} can be used to retrieve the arguments that were available to the panel function at that position. In this case, it can be called without arguments as \preformatted{ trellis.panelArgs() } This usage is also allowed when a \code{"trellis"} object is being printed, e.g. inside the panel functions or the axis function (but not inside the prepanel function). \code{trellis.panelArgs} can also retrieve the panel arguments from any \code{"trellis"} object. Note that for this usage, one needs to specify the \code{packet.number} (as described under the \code{panel} entry in \code{\link{xyplot}}) and not the position in the layout, because a layout determines the panel only \bold{after} the object has been printed. It is usually not necessary to call \code{trellis.vpname} and \code{trellis.grobname} directly. However, they can be useful in generating appropriate names in a portable way when using grid functions to interact with the plots directly, as described in the note below. } \value{ \code{panel.identify} returns an integer vector containing the subscripts of the identified points (see details above). The equivalent of \code{identify} with \code{pos=TRUE} is not yet implemented, but can be considered for addition if requested. \code{trellis.panelArgs} returns a named list of arguments that were available to the panel function for the chosen panel. \code{trellis.vpname} and \code{trellis.grobname} return character strings. \code{trellis.focus} has a meaningful return value only if it has been used to focus on a panel interactively, in which case the return value is a list with components \code{col} and \code{row} giving the column and row positions respectively of the chosen panel, unless the choice was cancelled (by a right click), in which case the return value is \code{NULL}. If click was outside a panel, both \code{col} and \code{row} are set to 0. } \note{ The viewports created by lattice is accessible to the user only up to a certain extent, as described above. In particular, \code{trellis.focus} can only be used to manipulate the last plot drawn. For full flexibility, use appropriate functions from the grid package directly. For example, \code{\link[grid:current.viewport]{current.vpTree}} can be used to inspect the current viewport tree and \code{\link[grid:viewports]{seekViewport}} or \code{\link[grid:viewports]{downViewport}} can be used to navigate to these viewports. For such usage, \code{trellis.vpname} and \code{trellis.grobname} (with a non-default \code{prefix} argument) provides a portable way to access the appropriate viewports and grobs by name. } \examples{ \dontrun{ xyplot(1:10 ~ 1:10) trellis.focus("panel", 1, 1) panel.identify() } xyplot(Petal.Length ~ Sepal.Length | Species, iris, layout = c(2, 2)) Sys.sleep(1) trellis.focus("panel", 1, 1) do.call("panel.lmline", trellis.panelArgs()) Sys.sleep(0.5) trellis.unfocus() trellis.focus("panel", 2, 1) do.call("panel.lmline", trellis.panelArgs()) Sys.sleep(0.5) trellis.unfocus() trellis.focus("panel", 1, 2) do.call("panel.lmline", trellis.panelArgs()) Sys.sleep(0.5) trellis.unfocus() ## choosing loess smoothing parameter p <- xyplot(dist ~ speed, cars) panel.loessresid <- function(x = panel.args$x, y = panel.args$y, span, panel.args = trellis.panelArgs()) { fm <- loess(y ~ x, span = span) xgrid <- do.breaks(current.panel.limits()$xlim, 50) ygrid <- predict(fm, newdata = data.frame(x = xgrid)) panel.lines(xgrid, ygrid) pred <- predict(fm) ## center residuals so that they fall inside panel resids <- y - pred + mean(y) fm.resid <- loess.smooth(x, resids, span = span) ##panel.points(x, resids, col = 1, pch = 4) panel.lines(fm.resid, col = 1) } spans <- c(0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8) update(p, index.cond = list(rep(1, length(spans)))) panel.locs <- trellis.currentLayout() i <- 1 for (row in 1:nrow(panel.locs)) for (column in 1:ncol(panel.locs)) if (panel.locs[row, column] > 0) { trellis.focus("panel", row = row, column = column, highlight = FALSE) panel.loessresid(span = spans[i]) grid::grid.text(paste("span = ", spans[i]), x = 0.25, y = 0.75, default.units = "npc") trellis.unfocus() i <- i + 1 } } \seealso{ \code{\link{identify}}, \code{\link{Lattice}}, \code{\link{print.trellis}}, \code{\link{trellis.currentLayout}}, \code{\link[grid:current.viewport]{current.vpTree}}, \code{\link[grid:viewports]{viewports}} } \author{ Deepayan Sarkar \email{Deepayan.Sarkar@R-project.org}. Felix Andrews provided initial implementations of \code{panel.identify.qqmath} and support for focusing on panels interctively. } \keyword{dplot}