NEWS.md
Small change to use gridGeometry 0.4.1 from Github until released to CRAN, this version fixes rare error caused by proportional diagram and innerborder, when the delta from two circles creates a small line with zero area. Will update again once it is available on CRAN. Thanks pmur002 for the rapid fix!
steps=20, formerly steps=200 which was (and is) the predominant rate-limiting step.nudge_JamPolygon() now recognizes rotate_degrees fixing the bug that previously did not allow rotating the Venn diagram polygons.
":" instead of ": " between the directional arrow and the count, for example "^^:20" instead of "^^: 20". Somehow the whitespace calculation is inconsistent, causing some labels to be wider than others with gridtext::richtext_grob(), with much wider whitespace than anticipated. The effect was inconsistent even on the same machine, between RStudio and R console, and differed across other machines also. Workarounds include choosing different fonts, apparently some provide more reliable whitespace calculations. The change only occurs during rendering when grobs are created. The underlying Venndir data is not changed.plot.JamPolygon(), affecting venndir() and render_venndir()
Now renders borders using gridGeometry thanks to suggestion from pmur002. Borders are more consistent, without small visual artifacts.
Added "outerborder" as a formal border type.
All three types of borders can be rendered, in order: outer, inner, border
border is not drawn, either the innerborder or outerborder are drawn on the border itself to prevent a tiny artifact gap between the innerborder and outerborder.Venndir object now uses outerborder instead of border, and sets border to NA.lwd controls default border line width.unicode (silent argument passed to curate_venn_labels() controls the Unicode character with up/down arrows, disagreement, etc. You can supply unicode=2 for alternate symbols, though they are font-dependent, R-dependent, and terminal-dependent. All things must work well together (apparently). Use unicode=FALSE for simple text.poly_alpha=0.6 makes background less intense.make_color_contrast() properly without forcing saturation, previously caused blue to become cyan instead of light blue.sample_JamPolygon() changed n_ratio=1 after testing and disliking the previous n_ratio=4.
venndir_to_df() gained some new features:
new argument df_format with three formats:
"hits" - essentially a hit matrix with 1, 0, -1 indicating direction"items" - each column of a data.frame contains items for a Venn overlap."wide" - intended as an RMarkdown summary - it uses grouped rows to display items in each Venn overlap. Not sure it works very well.argument return_type:
"data.frame" (default) - returns a data.frame"kable" - returns a colorized kable table for RMarkdown HTML"color" mainly used for venndir_to_df()
colorjam::col_linear_xf() for better colorization of counts when color_by_counts=TRUE.return_type="JamPolygon" since transition to JamPolygon, removed the option for "polygon_list".polygon_circles(), polygon_ellipses() both now return JamPolygon.
nudge_JamPolygon() - simple function to adjust JamPolygon polygonsCleaned up some generic function logic, probably more to do.
Venndir objects
JamPolygon objects either in a single list, multiple list, and will now retain all colnames across all JamPolygon objects.The polygon_list format loosely conformed to polyclip data input/output, but had several exceptions that motivated me to use JamPolygon: simple polygons encoded as list, complex polygons encoded as nested list; solid polygon encoded as clockwise points, holes encoded counterclockwise.
bbox_polygon_list()get_largest_polygon_list()intersect_polygon_list()labelr_polygon_list()minus_polygon_list()plot_polygon_list()polygon_list_labelr()rescale_polygon_list()union_polygon_list()get_venn_shapes()eulerr_to_polygon_list()polygon_areas()nudge_polygon_coords()nudge_polygon_list()polygon_list_to_xy_list()xy_list_to_polygon_list()spread=TRUE so that item labels are more evenly distributed by defaultn_ratio to control the target number of points, from which spread=TRUE will choose n evenly spaced. For now, the ratio is set n_ratio=5 which seems to perform well.... to sample_JamPolygon()
Change: names(setlist) will have any characters ":" changed to ".".
":" character is used by grid as a delimited to name grobs, which means names(setlist) that also contain ":" create confusion. Easiest workaround is to maintain ":" characters in the setlist_labels and legend_labels (for display) but to remove the ":" characters from the internal representation of these names.gridtext_richtext_grob() and supporting functions.render_thin_border to control whether the thin line between inner and outer border is drawn. It may be hidden by default in future.buffer can now be vectorized, to shift the plot along each side in order: (bottom, left, top, right). The default applies buffer to all side, but sometimes it is useful to shift the plot away from one side, for example buffer=c(0.1, 0, 0, 0, 0). It still maintains proper aspect ratio."" line colors, both are handled silently by drawing either line with width 0.01, or transparent color, respectively.expand_fraction to plot.JamPolygon using argument buffer so it can shift the Venn diagram off-center.... to plot.JamPolygon() so that optional arguments will be recognized.Substantial changes.
The method of assembling labels now arranges labels adjacent to each other, removing an important dependency on internal gridtext functions (which are not permitted in CRAN/Bioconductor). It also allows more flexible labeling in future, e.g. different overlap metrics, optional overlap names.
Labels use a more aesthetic layout: * The main set label is always top-center of a group of labels. * Count labels are split into left/right columns, for main/signed counts, respectively. They are middle-aligned left-to-right. * Set label and count labels are arranged top/bottom, also centered. * One region is defined for the group of labels. * The previous arrangement can be see with group_labels=FALSE.
All graphics are returned as grid graphical objects (grobs), which enables flexible options for drawing the figure. For example it can be drawn inside a grid::viewport, or included with patchwork or cowplot as a figure panel. Or, the grobs can be directly manipulated.
"gtree", "grob_list", "viewport" when do_plot=TRUE
padding and radius using the same values."gtree", "grob_list", "viewport"
grid::grid.draw() instead of being done piecemeal, making it appear faster.gridtext_richtext_grob(), since that also calls functions internal to the gridtext package. It was done in order to supply vector of different padding values.gridtext::richtext_grob() in a list.legend_padding to control the table cell size.gridExtra::tableGrob() calculation of padding caused left-aligned text to appear at the very left edge of the text box, without a padded region shown. The padding is only used to extend the box itself, but still allows the label to be aligned relative to the extended box, and not the inner region. This bug was sidestepped by enfocing leading/trailing whitespace.grid graphical objects, and the viewport defined.do_draw to control whether the figure is drawn, or only the graphical objects should be returned.draw_gridtext_groups() - substantial changes, most all previous processing has been changed. It is used internally.
Not terribly exciting, but useful nonetheless. Several new functions were added to help manipulate grid grobs, to stack, tile, xalign, yalign grobs relative to each other, using units that keep them aligned even when the plot device is resized. These functions are probably dependent upon gridtext labels, since they are used to create individual overlap and count labels, and define xext,yext with the “true” grid dimensions for each label. These functions make it possible to center the grouped label on the original label point, or align the left column with the right column, etc.
venndir() - relatively minor update, fixes unnamed setlist or setlist with duplicate or NA names.
names(setlist) are defined, otherwise integer numbers are defined in order.names(setlist) are unique, and not NA, otherwise it calls jamba::makeNames() to create unique names.Fixed bug in venndir() when using sets to define a subset from setlist causing the colors to be out of sync.
NA for names(setlist) are handled.Main goal is to fix bug associated with empty or nested setlist entries when used with proportional=TRUE. This update is somewhat of a risk, several core components were updated in venndir(), render_venndir() and venndir_label_style(), which could impact a number of downstream steps. Corresponding testthis entries have been added, hopefully to prevent occurrence of new bugs.
Added vdiffr to R package Suggests.
venndir(), render_venndir(), venndir_label_style()
Venndir slot jps with column "ref_polygon". This new column associates the set name with an overlap polygon, which points the set label to the appropriate part of a polygon. For example, by default “set_A” will point to the overlap_set “set_A” which has no overlaps with other sets. For proportional diagrams, sometimes “set_A” is fully encompassed by “set_B”, in which case the label “set_A” will point to the overlap polygon which is (1) not empty of course, and (2) has the fewest overlaps.overlap_set by a different name, for example "set_A&set_B". That change broke the label grouping logic, and needed to be resolved.testthis scenarios have been added, including the first examples that test consistent graphical output. Mainly these tests focus on proportional 2-, 3-, and 4-way Venn and Euler diagrams, with empty sets, or nested sets, or both.matrixStats from dependencies.setlist_labels and legend_labels that use names(setlist) by default, but can be adjusted to be display-friendly. The setlist_labels are displayed on the Venn diagram itself. The legend_labels are displayed only in the Venn legend via venndir_legender().list2im_opt() has a new default argument do_sparse=FALSE.
Matrix class when available. In almost all cases, I immediately converted back to matrix or used do_sparse=FALSE anyway. Before wider use of the package, best to improve the default condition.Added examples to rescale_coordinates().
venndir() and render_venndir()
show_labels which includes the letter "p". When "c" and "p" are both included, the two labels are included in the order they appear, separated using percent_delimiter which by default uses a newline. For now, percentage will always appear in the same location with counts "c", unless counts are hidden. For example show_labels="Ncp" will place set names outside, and count and percent inside. Similarly show_labels="Np" will place set names outside, and only the percent will be displayed inside. Also, show_labels="Npc" will place the percentage first, then the overlap count.item_style="default" will auto-detect Markdown and imposes item_style="gridtext" when found, otherwise uses the faster item_style="text".show_items=NA default is changed when items should be displayed, defined when show_labels contains "i". In that case the overlap_type helps determine eithr show_items="item" or show_items="sign item".Venndir-class - additional documentation.
percent_delim controls the delimiter between count and percent, and takes effect only when both are visible. The default "<br>" places the second value on a new line.venn_meme() change via hotfix:
draw_legend=FALSE to hide the Venn legend by default, a change from the default for venndir() and render_venndir(), since the legend is not relevant for venn_meme().sp and rgeos references were removed!Venndir objects internally, with “convenience functions” to help access and manipulate its data.add_orientation_JamPolygon(), area_JamPolygon(), plot.JamPolygon(), union_JamPolygon()
The polyclip update. Removing all remnants of sp and rgeos.
Major change, all functions ported or rewritten not to rely upon sp, rgeos, and sf, since the migration to sf would also incur heavy burden of installing sf and all its map/geography-related system libraries.
Backward compatibility is limited for pre-existing venndir output, since rgeos was removed from CRAN (ugh). In principle a pre-existing object could be re-run to produce the new object output.
Attempts were made for venndir() to produce output equivalent to previous versions, but it is unclear how practical that may be.
All graphical output uses grid, no support for base or ggplot2. The grid output was chosen since it can be conveniently combined into panels using patchwork or cowplot, while offering more robust features to position graphical components, also keeping 1:1 aspect ratio so Venn circles remain circular.
New S4 objects:
JamPolygon - makeshift polygon data.frame to hold one polygon per row (where one “polygon” may be represented as a multi-part polygon with optional holes or nested holes.) This object makes calling polyclip and grid functions more reliable, since they recognize slightly different formats.Venndir - replacement for list output, with equivalent content. Will transition to function accessors to obtain content, removing need for direct list access.pracma, polyclip, vwline to Depends.sf, sp, rgeos from Depends. (Ack)ggtext from Depends, all output uses gridtext or grid.venndir() and render_venndir() with new versions, very temporarily moving previous functions to venndir_OLD() and render_venndir_OLD().spsample().JamPolygon object and functions(It may eventually become its own R package.)
JamPolygon is intended to provide a very basic mechanism to store polygon coordinates with the following design:
data.frame.union_JamPolygon().)"evenodd" logic, so any inner polygon is assumed to be a hole, and any polygon inside a hole is assumed to be solid. This logic is applied regardless the clockwise/counterclockwise orientation of points for each polygon (polyclip convention: solid is clockwise, holes are counter-clockwise). Calls to polyclip functions will adjust the polygon orientation as needed for the purpose of using polyclip, but these details are not important for JamPolygon itself.innerborder is a new feature, my selfish desire to enable borders that are not overdrawn with adjacent polygons (which share the same border but on different sides). It requires a brilliant package vwline to draw variable-width lines (thanks Paul Murrell). It also requires knowledge of the direction lines are drawn, in order to define left/right border relative to the path taken when rendering the border.grid graphics. All coordinates are encoded in the JamPolygon object, so they can be plotted separately, however plot() and plot.JamPolygon() use grid.Some supporting functions for JamPolygon:
plot.JamPolygon() - grid draw function for JamPolygon objectsrbind2() - generic function to combine multiple JamPolygon objectsbbox_JamPolygon() - bounding box enclosing all polygonsarea_JamPolygon() - area for each row, accounting for holes and nested polygons.buffer_JamPolygon() - expand or shrink polygon borderspoint_in_JamPolygon() - test for point inside solid portion of polygonshas_point_in_JamPolygon() - logical presence/absence of point in any solid polygonadd_orientation_JamPolygon() - adds polygon/multipolygon orientation, including holes, clockwise, and parent enclosing polygon.labelr_JamPolygon() - define best label position within each polygonminus_JamPolygon() - first polygon, removing subsequent polygonsunion_JamPolygon() - combine and simplify all polygons into one rowintersect_JamPolygon() - first polygon, intersecting subsequent polygonssample_JamPolygon() - define sample points within solid polygonsplit_JamPolygon() - split multi-part polygons onto separate rows. Note that holes become solid polygons during this step.polyclip_to_JamPolygon() - convert polyclip::polyclip() output to JamPolygon.sp to polyclip, avoiding heavy dependencies on GDAL, GEOS, LWGEOM software libraries which are unrelated to venndir’s use of polygon geometry functions.polygon_label_fill() threw an error when supplied with an empty polygon, now returns NULL so the upstream function can deal with the lack of labels. The issue appears to arise from proportional diagrams that use item labels, where the item label has no suitable polygon to use. Presumably something from the updated eulerr package that creates overlap circles/ellipses to use in these diagrams.reposition_venn_gridtext_labels() intended to help adjust and re-position gridtext labels relative to optional incoming line segments, and with goal of enabling different layouts for combinations of labels.character strings. For example from="1" will no longer match the value "-1" when it is the first entry of curate_df[,"from"]. Similarly, an output string that contains "1" will no longer be re-replaced again by matching pattern from="1". The use case was noticed when trying to specify a gridtext font size using from=c("-1", "1") and sign=c("<span style='font-size=18pt'>Down</span>", "<span style='font-size=18pt'>Up</span>") which caused the "1" to be re-replaced.plot_style="base" now plots the sf form of data, rather than SpatialPolygonsDataFrame, the first step in transitioning from sp to sf prior to the deprecation of the sp package altogether. All plotting should be equivalent.lwd is now honored when plotting Venn circles and polygons, and lwd=0 is silently replaced with lwd=1 and col="#00000000" so the borders are hidden.plot_style="gg" honors lwd as linewidth using ggplot2::scale_linewidth_identity(), which unfortunately is slightly larger line width than corresponding values with base R plots. Ah well.lwd=2 to lwd=0.5. The value is not configurable in function calls, but can be edited in the venndir output list object, in venn_spdf.setlist and order of the venndir_out, in case using a subset of setlist in venndir(..., sets=c(1, 3, 4))
venndir_out was supplied, now it defaults to white background with black text.keep_newlines=FALSE which by default removes newlines (line breaks) in the set labels. Use keep_newlines=TRUE to keep line breaks as-is.set_colors to specify optional colors to use in the table.poly_alpha value used to define transparency of Venn colors is honored in the color legend, using venn_spdf$alpha."border" is now honored, as is "lwd".venndir_legend()
style="grid" as it offers a nicely organized style, with better positioning which works the same for base and ggplot2 venndir() output.data.table to dependencies, since it is used by shrink_df().gridExtra to enable optional legend gridExtra::tableGrob().overlap_type="concordance" was changed to overlap_type="detect" to fix regression where all input data is assumed to be signed, thus showing up arrows even for non-directional input data. The function signed_overlaps() will detect an appropriate type unless given a specific overlap_type value.list2im_opt() and list2im_value()
do_sparse=TRUE was changed to do_sparse=FALSE because in almost all cases the most appropriate class is matrix. Only for very rare cases with extremely large matrices should sparse matrix objects be used, which makes it most appropriate as an option and not a default.counts2setlist() warnings regarding GEOS future retirement are silenced in the examples.list2im_opt() uses the updated non-deprecated coersion of matrix to "ngCMatrix", although the end result is still an object with class "ngCMatrix". It now uses three as() coersions, as instructed by the Matrix package. The Matrix as() is properly imported by using importFrom(methods, as) and placing Matrix into the Suggests list.jamba now requires version 0.0.90.900 for coordPresets() to provide figure positioning rather than plot positioning. Used for plot_warning with proportional Venn diagrams that cannot display all possible overlaps.padding to customize the padding around Venn labels.padding is scaled relative to font_cex, for smaller padding when the font size is scaled down.plot_warning label is placed relative to the bottom of the plot figure, not the plot border, giving a bit more space below the typical proportional Venn diagram.render_venndir() calls gridtext_richtext_grob() instead of gridtext::richtext_grob() to enable vectorized padding.
Two temporary functions to customize gridtext. Both functions also call internel gridtext functions, so they cannot be maintained when venndir is prepared for CRAN. More likely these functions will become the basis for gridtext issue and/or pull request (PR).
gridtext_richtext_grob() - apply padding in vectorized form, to enable distinct padding to be applied to each label. This function is modified to extend padding and call custom gridtext_make_outer_box().gridtext_make_outer_box() - apply padding in vectorized form, to enable distinct padding to be applied to each label.venndir_label_style() position for set names is now "outside", changed from previous "inside". There were too many cases where the set name was long, then pushed the numeric count labels too far away from the desired position at the center of the Venn polygon. The new challenge is getting adequate spacing around the Venn diagram to place labels that fit inside the plot window.make_color_contrast() was adjusted to reduce the saturation/chroma of light text colors, the end result reduces the “cyan” appearance of “dodgerblue” when converted to a lighter contrasting color.venndir() new argument sign_count_delim allows customization of the delimiter between sign and numeric counts, for example "^^: 12" uses the default delimiter ": ". In some cases it may be preferred to use sign_count_delim=" " to omit the colon ":" character.venndir() and related functions were failing in RStudio when used under devtools::load_all(), apparently because the rgeos package needs to be loaded with @imports before directly calling functions from the rgeos package. This issue appears to occur only when using devtools.
Corrected some warnings caused by argument overloading:
label_preset is now recognized by venndir() and render_venndir(), and is now passed formally from venndir() to render_venndir() to prevent it being passed to other internal functions, such as the plot() function. The warning was: "label_preset" is not a graphical parameter
display_counts a developmental argument passed from venndir() to render_venndir(). The warning was: "display_counts" is not a graphical parameter
"sp-to-sf" to migrate away from the sp (Spatial Polygons) package, and toward the sf (Simple Features) package. This migration will remove dependencies on sp and rgeos which are being retired in 2023 anyway. It should also make the overall dependency footprint of venndir much smaller, since sf has few dependencies while also covering the functions required from other packages.textvenn() was throwing an error with jamba version 0.0.83.900 caused by updates to jamba::printDebug() used for printing colorized output. Dependency was bumped to 0.0.84.900 where the bug was fixed.venndir() and render_venndir() new argument item_buffer
polygon_label_fill() as argument scale_width which was otherwise very hard to find.render_venndir() will now auto-calculate an appropriate item_cex
item_cex=NULL or item_cex has only one numeric value. The scaling is based upon the total number of items inside each polygon, and the relative area of each polygon.item_cex is supplied as a single numeric value, it is multipled by the auto-scaled adjustment, so the item_cex acts as a relative adjustment to the auto-scale font sizing.item_cex_factor helps control the magnitude of scaling for large numbers of items (which shouldn’t matter much since they are already not very legible.).render_venndir() default change segment_buffer=-0.05
segment_buffer=-0.2
venn_meme() was updated to pass item_cex as a two-item numeric vector in order to prevent the auto-scaling described above.
Proportional Venn diagrams have been a bit of a challenge for labeling. Some edge cases were resolved finally, mostly relevant when certain Venn overlap wedges were razor thin, or when an overlap was represented by more than one polygon in the resulting geometry.
Also, label placement outside the Venn circle was inconsistent, some of the options in polygon_label_outside() appear to work poorly together. Its default values were adjusted to be more consistently “pleasing”.
render_venndir() had a bug that caused proportional Venn labels for single-set overlaps to be assigned to the wrong set polygon when the label was being positioned outside. The line segment would sometimes be visibly ambiguous. It would point the setA line segment to the full setA instead of the smaller non-overlap polygon representing unique setA items. In most cases the workaround was to use segment_buffer=0 so the label arrow is drawn to the edge of the circle. When the overall Venn diagram had reasonable proportions the visual effect was not that obvious. With extreme Venn diagrams (small sliver of unique items for some sets) the output was not as intended.get_largest_polygon() was updated to work with SpatialPolygonsDataFrame, previously it did not subset the polygons, returning the full set of polygons. It contributed to set labels being weirdly placed, specifically in rare cases where an overlap set (for example setA&setB) was represented by two polygons where setA and setB uniquely overlap, and where one of those polygons was extremely small. In the weird test case, the smaller polygon had area 0.2, while the larger polygon had area 150. The label is intended to point to the largest available polygon for each overlap, however with objects SpatialPolygonsDataFrame it was returning all polygons for each overlap, and the first polygon was labeled. The issue showed up only when the first of multiple polygons happened to be the smaller one, and substantially smaller.Lightweight change: polygon_label_outside() was moved into a separate .R file. The venndir-sp.R file has too many functions, it will be split into pieces eventually.
polygon_label_fill(): At last fixed an elusive bug causing certain item label scenarios to throw an error, turned out to be caused by lack of rownames on a matrix sorted by jamba::mixedSortDF(). The error occurred in an internal function get_poly_points() that iterates different labeling conditions until at least n label coordinates are returned. Apparently the function spsample() does not guarantee returning the same number of coordinate points as requested. This error must have bitten someone else somewhere in the R world, requesting 15 sampled points, then receiving only 12.rescale_sp() argument rotate_degrees will now recognized a named vector.
sp object in the ID slotName of each individual Polygons object, for example the first polygon in a SpatialPolygon object would have name stored as: sp@polygons[[1]]@ID. When rotate_degrees contains names, only names that match the polygons will be applied, which allows rotating individual polygons within a SpatialPolygons object.share_polygon_center=TRUE new argument intended for multi-polygon Polygons objects, which happens when there are two disconnected parts inside one Polygons element. When this occurs, by default the two parts are rotated as if they were one unit, around their own local center, by default. When share_polygon_center=FALSE each polygon is rotated about its own personal center.item_style="gridtext" is a different default than venndir() because memes are expected to have fewer labels, and are most likely to use fancy markdown text formatting. This argument is passed to venndir(). When item_style="text" it will autocorrect "<br>" into newline; when item_style="gridtext" it will autocorrect newline into "<br>".To resolve issue #2, the colorjam package was updated to add prefix to call jamba::tcount() to function colorjam::approx_degrees(). When the colorjam package is loaded, using library(colorjam) or require(colorjam) it automatically also loads jamba so its functions are available. However, when colorjam::approx_degrees() is called directly, without formally loading the colorjam package, the jamba functions are not loaded. This issue caused the error, which unfortunately causes the venndir() function to fail unless jamba or colorjam packages have been loaded previously.
All that said, the “fix” is to require the updated colorjam(>= 0.0.22.950) package version that includes this fix.
venndir() was updated to change the priority of label placement for proportional Venn diagrams. It used to choose the full set circle for the main set label, now it chooses the unique overlap polygon for that set if it exists, then uses the full set circle only as a fallback.venndir_label_style() was updated to fix longstanding weird glitch with proportional Venn diagrams sometimes displaying counts when one set is fully inside all other sets, but that set has non-zero items unique to that set - so there is nowhere appropriate to display them. Because the main set label was displayed, the count label was errantly placed at the center of the Venn circle. Now the bug is fixed, the value is not displayed - it is part of the optional warning message that lists any overlap counts that could not be displayed.render_venndir() fixed type “bow_lwd” is now correctly “box_lwd”, only impacted box line width when the line width was changed from default.counts2setlist() corrected an error when supplied with empty overlaps, which previously included at least one delimiter even for empty sets. Entries were added to the tests to confirm this behavior.polygon_label_fill() small bug in how label placement used rotation, the polygon center was not properly re-used when un-rotating labels to the original orientation. The default was -20 degrees, so the effect was subtle.venndir_label_style() was updated to change the logic used to place labels inside or outside each Venn overlap polygon, specifically to allow item labels to be inside, or hidden, and adjust the count label visibility accordingly.
venndir_label_style(), and venndir() now properly hide item labels using max_items, instead displaying the overlap count.
venndir_label_style() arguments count and signed have new options:
"ifneeded" will display the overlap count when item labels are not being displayed, otherwise when item labels are displayed the count will be hidden."detect" will display the overlap count inside when item labels are not being displayed, otherwise the overlap count is outside when item labels are displayed.polygon_label_fill() new argument ref_sp used to know the relative size of the polygon being labeled, to account for the adjustment when apply_n_scale=TRUE: the buffer polygon is generally smaller for fewer labels, but the new adjustment prevents polygons from being shrunk when the polygon is already very small.
signed_overlaps() new default overlap_type="default" which attempts to choose a reasonable option. When all values are c(0, 1) it will choose overlap_type="overlap" which does not use directionality. When the input contains numeric or integer values it will enforce sign() so the comparison is done with the sign and not decimal values by default.
signed_overlaps() was not handling factor vectors as input, they were getting coerced to integer values instead of character values. This issue was caused by handling duplicate entries, in this case the duplicate entries were not renamed to unique entries by calling c() for the makeNamesFunc part of jamba::makeNames(). The base R assignment names(x) <- c(y) coerces factor values in y to integer values, which in this case breaks the expected behavior. This seems like an edge case, but certainly one that venndir should handle without issue - it is not always clear when values are stored as factors instead of character, and the expected behavior in all cases is for venndir to treat factor as character values. (Thank you Dr. Ashley Brooks for bringing my attention to this issue!)list2im_value() was also updated to handle factor values by coercing to character instead of allowing the base R matrix coersion which by default coerces to integer.list2im_value() was also updated to handle empty which is a user-defined value to use in the resulting incidence matrix for empty entries. This argument is not used by venndir but is available for general conversion of value list to value incidence matrix.list2im_value() was also updated to use a consistent default empty based upon whether the input setlist contains any character or factor values. The previous behavior handled each column independently. The rule is that when any column contains character or factor values, the default is empty="" across all columns; if all values are numeric the default is empty=0. The result is a consistently formatted value incidence matrix. This is a rare edge case that is not expected for common use of venndir but may arise from the conversion functions from value list to value incidence matrix.signed_overlaps() with named factor vectors as input.venn_meme() is a wrapper function to help create “Venn Memes” also known as “Concept Venns”. These have text inside each overlap but no other counts or labels. Input is expected to be an overlap list, which is converted to setlist using overlaplist2setlist(). See examples.venndir() argument font_cex was updated to accept length 3 which is applied to (1) main set labels, (2) main count labels, (3) directional count labels. It is sometimes useful to have main set labels larger than the main count labels, which prompted this change. Now it is possible to place main set labels outside the diagram, and have smaller count labels inside the diagram, for example to fit inside proportional Euler diagram shapes. This change adds fontsize to the venn_spdf output object, thus storing the set label font size separate from the count label font size.venndir_label_style() new argument option label_preset="custom" will not modify any label position settings. This option is intended when labels are manually adjusted and should not be changed, and where the user wants to apply label_style to affect the label color, fill, and box outline.colorjam to version 0.0.18.900 to fix error with rad2deg() when blending colors.venndir_label_style() now recognizes when items cannot be displayed inside proportional diagrams, and sets show_items="none" for those entries in label_df. This change prevents trying to place labels inside a NULL overlap polygon.venndir converters like im2list_opt(), list2im(), im_value2list(), and list2im_value().label_polygon_fill() new argument layout_degrees allows rotating the label placement, relevant mostly for label_method="regular" which produces square point grid, and label_method="hexagonal" which produces a hexagonal point grid. It can be useful to rotate those points so text labels have less overlap. As a result degrees=0 is the new default, as layout_degrees=-22 is also the new default. See the new set of examples which includes visual representation of methods to reduce text label overlaps.render_venndir() default is changed to item_degrees=0.match_list() was updated to require both list elements to have the same length, to account for duplicate items.label_polygon_fill() to polygon_label_fill() to be consistent with polygon_label_outside() and polygon_label_segment() – also before anyone happens to start using this function!fontfamily was not being used with item labels, breaking display of unicode characters in some sessions.to_basic.GeomRichText() and to_basic.GeomTextBox() were added for rudimentary compatibility when using plotly::ggplotly() with ggtext::geom_richtext() and ggtext::geom_textbox(). These functions do display text inside plotly figure, however: (1) they do not cleanly use text positioning, even with textposition;render_venndir() via plot_style="plotly" then using plotly::plot_ly() commands in native form.show_zero=TRUE was not working, zeros were always hidden.alpha_by_counts appears to be mis-aligned.make_color_contrast() without using the correct alpha value. See with venndir(make_venn_test(100, 3), poly_alpha=0)
render_venndir() by default groups gridtext labels together, and draws one contiguous background and border around them all. It currently only works for R base plots, since it needs to know the grob sizes upfront. That said, at this point the base R plot looks substantially better than ggplot2 when labels are placed on the outside of the Venn circles. Hopefully I will figure out how to do the same for ggplot2 output.draw_gridtext_groups() does the work for grouping multiple gridtext::richtext_grob() labels together, using x, y, overlap_set to define the grouping. This method also aligns text relative to a line segment if present, by using degrees_to_adj(). A by-product of the approach is that labels are now centered by the label width, instead of having left/right components which may differ in width.Began adding testthat tests for various functions.
Proportional Venn diagrams (Euler diagrams) warranted a refactor in order to label each set even when one set is fully contained inside another set.
signed_counts2setlist() is an import method similar to counts2setlist() but to allow importing directional overlaps.get_sp_buffer() applies a buffer width to the interior of a polygon, either with fixed width, or relative width based upon the minimum width required to make the polygon disappear. The relative value means sp_buffer=0.5 can be used to create a buffer at half-width, for a polygon of any size. The purpose is to draw a line segment from outside the polygon to “just inside” the polygon, using an amount appropriate for the size and shape of the polygon. As “rgeos” gBuffer() has no ability to scale relative to the polygon size and shape, get_sp_buffer() quickly runs a sweep to determine the width where an internal polygon is just barely non-zero, and defines that as sp_buffer=-1.mean_degrees() takes a vector of angles in degrees and returns the mean angle based upon the sum of unit vectors. It adds a small random bit to prevent vectors from exactly cancelling out, and an argument seed that makes the randomness reproducible.mean_degree_arc() is similar to mean_degrees() except it takes an ordered set of angles that define an arc, and returns the mean position along that arc. Use argument do_plot=TRUE to see a visual representation of the options. It turns out that spreading text labels along an arc only worked when the mean angle was fixed inside the arc. Otherwise occassionally the mean angle could be exactly 180 degrees off, when the angles span more than 180 degrees, and if multiple labels were present at each end. You’d think that would be rare, but it happened enough to notice the bug and to implement better logic.degrees_to_adj() is a helper function that takes an angle in degrees, and returns the text adjx and adjy values so a text label will be positioned outside the angle.ggrender_venndir() was removed, use render_venndir(plot_style="gg") in its place. It made sense to keep the visualization logic inside one function.spread_degrees() was not properly maintaining the initial order of angles when it iteratively spread out a series of angles. For example one group of angles could be spread wide enough to overlap another existing angle, in which case that angle should be kept outside the initial group, instead of being placed into the middle. The telltale sign is that line segments wildly criss-crossed inside the polygon.New argument inside_percent_threshold added to venndir(), render_venndir(), and venndir_label_style() used to position labels outside a polygon when its size is below this percent of the total polygon size. It works as a reasonable alternative for determining if the label visually fits inside a polygon.
venndir() can accept polygon shapes to use as input, instead of defining its own or using those from eulerr::euler(). New argument venn_sp accepts any SpatialPolygons compatible object, which must have at least as many polygons as there are sets in setlist.
venndir() now adds the Venn shapes to the output venn_spdf with type="set"; all other polygons have type="overlap". The shapes will serve as an anchor for set labels, in the event we want to label one or more circles – like when one circle is fully contained inside another.
polygon_label_outside() now determines a reasonable text alignment adj based upon the angle offset, to help minimize label overlap with the line segment.
polygon_label_outside() was updated to correct some internal use of coordinates. New argument sp_buffer to allow the line segment to end slightly inside each polygon.
get_venn_shapes() default 4-way Venn ellipses were slightly adjusted so the middle ellipses broke the outer ellipses into two polygons. This change helps the automated position of labels by polygon_label_outside() with certain settings.
venndir_label_style() was refactored to be able to position labels inside or outside the polygon, for four types of labels:
"set" - e.g. "set_A"
"overlap" - e.g. "set_A&set_B" normally this label is not shown
"count" - e.g. 10
"items" - e.g. c("item1", "item3", "item9") for each overlap set.
It has some label_preset options, which help position set and count labels inside or outside.
One more refactor is in progress. Label coordinates will be pre-calculated inside and outside each polygon, then labels can be toggled in or out (or moved to a custom location). The main driver is to be able to label the Venn circles on the perimeter of the diagram. For proportional circles, when a set is fully contained inside another set, there is no polygon with that set alone – thus the defaul labeling fails.
polygon_label_outside() defines label placement outside a polygon, one of two methods: “label” positions outside using relative label positioning; “bbox” positions outside using relative polygon position within the bounding box.expand_range() is a simple function to expand a numeric range by expand_fraction, mainly for R plots to have space to render labels outside a central Venn diagram.spread_degrees() takes a vector of angles in degrees as input, and returns a vector where angles are separated by at least min_degrees.mean_degrees() calculates the mean angle in degrees for an input vector of angles.diff_degrees() takes a vector of angles in degrees, and returns the angular difference. Suitable for cases with angles like c(355, 2, 12) where the difference between 355 and 2 is 7.sp_polylabelr() is a simple wrapper function to polylabelr::poi() for SpatialPolygons objects.Several refactoring processes being implemented, some further updates ad bugfixes may be necessary. Pushed this version for more testing on CentOS where the gridtext package was hard to install for GCC below version 5.3, but was installed with GCC 5.3.
Added dependencies:
sf package dependency for ggplot2 outputgridtext and ggtext for label outputgridBase for base R plottingrender_venndir() added sp prefix to plot() for base R plotting.signed_overlaps() was updated to allow using sep="|". Previous this delimited was used to make unique rownames, and needed to be parsed properly to prevent errors.venndir() new argument sep which is now properly passed to signed_overlaps() and get_venn_shapes().find_vennpoly_overlaps(), overlaplist2setlist(), counts2setlist() now call strsplit(..., fixed=TRUE) so the sep character is not treated like a regular expression, thus allowing delimiter sep="|" if specified.textvenn() new argument sep to allow customizing the delimiter between set names, for example the default sep="&" creates overlap names like set_A&set_B; but sep=" | " creates overlap names like set_A | set_B.
venndir() new argument show_set to customize display of each set name.
show_set="main" displays only the set name for main setsshow_set="all" displays all set names in each overlapshow_set="none" displays no set nameshow_set="boundary" not yet implemented will display the set label for each circle/ellipse boundaryMuch closer to a release-ready form. It needs the ggplot2 option enabled with the newest render_venndir() logic.
venndir() default show_zero=FALSE to hide empty labels.ggrender_venndir() from the exported functions list, this function needs to be included inside render_venndir().More refactoring may take place to simplify the workflow, until the package is publicized.
label_df$col renamed to label_df$color for consistency, and preparing for use by others.ggrender_vendir() will be retired, or will be converted to a simple wrapper function which calls render_venndir().label_polygon_fill() new arguments dither_cex and dither_color which slightly adjust the cex and color values. The cex is uniformly distributed +/- the dither_cex value, as a fraction of the cex, the default is roughly 4% variation in cex. The dither_color is similar, it adds some heterogeneity to darkFactor and sFactor and calls jamba::makeColorDarker() to adjust brightness and saturation.label_polygon_fill() can plot the polygon buffer behind the labels, to review the effect of adjusting scale_width for each polygon.label_df uses show_label and show_items to control visibility of the count label, and item labels, respectively. When the values are NA it applied default rules, which favors showing count label, unless the count label is moved outside the polygon.label_df recognizes x_offset,y_offset to nudge count labels somewhere else. Currently manual, this adjustment may be automated at some point. When a label is moved outside its parent polygon, a line segment is drawn from the label to just inside the polygon boundary. When no items are displayed, the segment is drawn more toward the center.venndir_label_style() new function to control the visual style of count labels, with option to fill and draw a border around labels.polygon_label_segment() is used when moving a label from inside a polygon to somewhere that may or may not be inside that polygon. When it is moved outside, it returns the point on the polygon boundary along that line segment. It can be used with sp_buffer so the line can be drawn slightly inside the polygon boundary.ggrender_venndir() and render_venndir() both allow optional display of items inside the Venn polygons. For relatively small sets (<500) this works fairly well. You can configure the label to include the "sign", and/or "item". It works as an alternative or supplement to using proportional Venn diagrams.ggrender_venndir() renders a venndir diagram using ggplot2. Note that this function requires the package sf, and in future other objects internal to venndir may be converted from sp to sf.label_polyogn_fill() takes a SpatialPolygons object, a vector of labels, and returns coordinates to place labels inside the polygon. It has some adjustments to add buffer around polygon boundaries, but does not directly detect label overlaps. It either plots labels in base R graphics, or returns coordinates suitable for ggplot2.rescale_*() functionsThe rescale_*() functions are intended to allow manipulation of numeric coordinates in each relevant object type, aimed at SpatialPolygons and the various sub-types.
rescale_sp() manipulates SpatialPolygons which contains one or more Polygons objects, and therefore calls:rescale_ps() manipulates Polygons which contains one or more Polygon objects, and therefore calls:rescale_p() manipulates Polygon which contains numeric matrix coordinates, and therefore calls:rescale_coordinates() manipulates numeric matrix coordinates.The rescale_*() operations enable three types of manipulations:
scale - multiplies coordinates for each axisrotate_degrees - rotates points in degreesshift - shifts coordinates by adding a constant to each axiscenter - the scale and rotate_degree are applied relative to the center, which by default uses the mean bounding box.list2im_opt() - optimized method to convert a list of character vectors to a logical incidence matrix. The list names become colnames, which represent the names of each set. The rownames represent items.list2im_value() - convert list of named vectors to a value incidence matrix. The list vectors are named by item, and the vector values are stored in the matrix.im2list() - convert incidence matrix to list of character vectors.im_value2list() - convert value incidence matrix to a list of vectors. Vectors are named by item, and values are from the incidence matrix.counts2setlist() - convert vector of overlap counts to setlist, intended for compatibility with other packages like eulerr, upsetr, and VennDiagram.venndir() new argument alpha_by_counts adjusts alpha transparency of each overlap based upon the number of counts. Implemented to evaluate the visual effect for now.venndir() no longer renders the plot, but calls render_venndir() in order to offer more customized options. It also allows editing the source data as needed, like adjusting individual overlap settings.textvenn() now supports directional Venn counts, even with "each" it uses two columns for all possible directions for the central 3-way overlaps.render_venndir() renders a Venn or Euler diagram using the output format from venndir(). It displays a warning at the bottom of the plot whenever a Venn overlap cannot be displayed, for example with proportional Venn (Euler) diagrams when an overlap cannot be represented. This label can be hidden, but by default is displayed so the information is not lost.venndir() the core visual function that displays either Venn diagram or a Euler diagram for proportional diagrams.The venndir package is intended for analyses where direction is encoded alongside each item in a set. The driving example is for gene expression analysis, which results in a subset of differentially expressed genes, where the expression can either by “up-regulation” or “down-regulation.” When comparing two sets of differentially expressed genes, it is helpful if not essential to recognize when any overlaps also share the same direction of change (or opposite direction.)
Although there are some 40,000 gene loci in human (give or take), it is common to see many of the same few hundred genes differentially regulated across a very diverse array of experiments. For two experiments to identify 800 genes each, with overlap of 400, these outcomes illustrate the utility of venndir:
textvenn() prints a Venn diagram using console text output. It also supports directionality, the formatting is still being adjusted for best visual appeal.signed_overlaps() is the core overlap function for venndir, it performs directional and direct overlap, with configurable output. It returns a data.frame with counts for each Venn overlap set, and optionally includes items. The exact output format may be adjusted in future, currently it returns one row per overlap. For directional overlaps, it returns one row per combination of up/down.make_venn_test() is a convenience function to generate test sets for Venn diagrams. It can return simple sets, or sets with directionality. It is mainly intended to help create very large sets to test efficiency of the process.