coord calendar

[cynthiahqy]

Use a wrapped cartesian coordinate system to plot temporal data in calendar like layouts instead of calculated variables (e.g. https://github.com/cynthiahqy/ggtilecal)

Should support:

• week in year layout with month line breaks;
• faceted months, years
• position after each wrap point -- 0 for over plot, 1, -1 for direction
• ragged layout
• split rendering of geometries across rows (maybe not possible in ggplot2)
• most other geoms out of the box (e.g. moon phases which calendR has options for)

Needs:

• Wrap point helpers similar to lubridate::floor_date() Gregorian calendar
• hack grouping to fix rendering limitations (for portals!)

Stretch goal -- wrapping non-gregorian calendar layouts:

• needs different datetime rounding system:
  • helpers for specifying "floor" functions e.g. floor_date_lunar()
  • return the nearest wrapping points
  • wrapping points via sequence generator OR lookup table

[mitchelloharawild]

Possibly needs two 'limits' - one for limits applied to continuous time (what the associated scale expects), and another on wrapped time (what the x-axis is after mapping from cartesian to the calendar coordinate space). This nuance could prove tricky to implement.

[cynthiahqy]

There might be some utility in having a time aesthetic, which defaults to x if unavailable. It frees up the x & y channels for things pie charts for daily data, or side by side columns like:

If we can use the "y"-axis to nest both data & time labels, then why not also the "x"-axis ?

Maybe also some utility in exploring calendar layouts using nested facet for similar reasons (freeing up the x/y channels). Personal data/storytelling scale lends itself to facet solutions, while big data/analytics/modelling lends itself more to coordinate solutions?

[mitchelloharawild]

Pie charts are a bit trickier than bar charts since there is a conflict in coordinate systems. The layout of pie charts respects a cartesian/calendar coordinate layout in time while the layout of x/y is in polar coordinates. Typical calendars are built on a cartesian coordinate system, and I'm not sure what a calendar built on a polar coordinate system would look like.

An idea for a calendar in polar coordinates could be:

• Each 'wrapped' breakpoint needs to cover a complete rotation of the polar coordinate system (justified length/duration)
• The calendar row is no longer a discrete jump to the next series, but instead perhaps a gradual increase over the wrapped time period.

Perhaps this is what a calendar looks like when built upon a polar coordinate system. The y-axis on the coordinate system is comprised within the black bands shown below.

library(ggplot2)
data.frame(
  y = 1:120,
  x = rep(1:12, 10),
  year = rep(2015:2024, each = 12)
) |> 
  ggplot(aes(x = x, y = y, group = year)) + 
  geom_line(linewidth = 5) + 
  coord_polar()

Or altenatively you can transform the calendar y-axis spacing to de-emphasise older observations in time.

data.frame(
  y = 1:120,
  x = rep(1:12, 10),
  year = rep(2015:2024, each = 12)
) |> 
  ggplot(aes(x = x, y = y, group = year, linewidth = y)) + 
  geom_line() + 
  coord_polar() + 
  scale_linewidth(range = c(1, 10)) +
  scale_y_continuous(transform = scales::transform_boxcox(2))

For example…

data.frame(
  y = 1:120 + runif(120, -6, 6),
  x = rep(1:12, 10),
  year = rep(2015:2024, each = 12)
) |> 
  ggplot(aes(x = x, y = y, group = year)) + 
  geom_line() + 
  coord_polar()

data.frame(
  y = 11:130 + runif(120, -6, 6),
  x = rep(1:12, 10),
  year = rep(2015:2024, each = 12)
) |> 
  ggplot(aes(x = x, y = y, group = year)) + 
  geom_line() + 
  coord_polar() + 
  scale_y_continuous(transform = scales::transform_boxcox(2))

^{Created on 2024-04-10 with reprex v2.0.2}