Major update to plot_ly()

[cpsievert]

@jackparmer @timelyportfolio

This is a major update to the plot_ly() interface that addresses a number of concerns raised by @hadley / @jcheng5 / others. See the NEWS file for a summary. My goal is to have this on CRAN before I fly out to SF on the 23rd. Here is a list of stuff that should be addressed before submitting:

• Get a working version of add_polygon()/add_ribbon(). BTW, if anyone has ideas for other add_*() functions (to do things that aren't easy/available or slow in ggplot2) I'm all ears!
• Add tests/documentation
• Fix grouping when there is multiple traces, for instance:

g <- expand.grid(visit = 1:10, id = 1:50, cohort = c("A", "B"))
g$response <- rnorm(nrow(g))
g %>%
  group_by(id) %>%
  plot_ly(x = ~visit, y = ~response, color = ~cohort, opacity = 0.5)

I also think that color should be deprecated and split into stroke/fill, but that should probably wait for now.

[cpsievert]

@timelyportfolio it looks like resetting zoom within an R markdown doc is broken on this branch. Could you check out #627 for me and see if it's just v1.13.0 that breaks it?

[timelyportfolio]

This will be really nice. I am working through examples and tests now. The first issue I noticed is in the ?plot_ly.

data(economics, package = "ggplot2")
# basic time-series plot
p <- plot_ly(economics, x = ~date, y = ~uempmed)
# add a loess smoother
p2 <- add_trace(p, y = ~fitted(loess(uempmed ~ as.numeric(date))))
# add a title
p3 <- layout(p2, title = "Median duration of unemployment (in weeks)")
# change the font
layout(p3, font = list(family = "Courier New, monospace"))

and I get

I would expect two lines (one raw and a second smoother). Also, the title is lost.

I'll work through, but I just want to report, so I don't forget.

[timelyportfolio]

this works

plot_ly() %>%
  add_data(economics) %>%
  add_trace(x = ~date, y = ~pce) %>%
  add_trace(x = ~date, y = ~log(pce,base=10)
)

so I'm struggling to understand why the other does not.

[timelyportfolio]

it seems this explains #628 (comment), but I don't understand the logic. Could you elaborate on this design?

[timelyportfolio]

If I comment out, then both the smoother and raw line are plotted, so this does seem to explain what is happening, but I don't understand why this choice was made. I'm guessing there are use cases where this is beneficial.

[cpsievert]

@timelyportfolio similar to ggvis()/ggplot(), I'd like to move to a system where plot_ly() initializes a visualization, and defines global attributes/properties, but doesn't necessarily define a trace (i.e., layer) itself. That way, this would create a single trace, rather than 2:

plot_ly(economics, x = ~date, y = ~pop) %>% add_lines()

If no add_trace()-like functions are used, we essentially add it at print time (so that these would also create the same result).

plot_ly(economics, x = ~date, y = ~pop, type = "scatter")
plot_ly(economics, x = ~date, y = ~pop)
No trace type specified. Using the 'scatter' default.

Unfortunately, this does introduce some confusing (backwards-incompatible) changes. Take the first example on plot_ly(). In the past, this would create two traces:

plot_ly(economics, x = ~date, y = ~uempmed) %>% 
  add_trace(y = ~fitted(loess(uempmed ~ as.numeric(date))))

But, given this new philosophy (which I think is better and more consistent with grammar of graphics systems), we'd expect that to create one trace. This is now the preferred way to express that plot:

plot_ly(economics, x = ~date) %>% 
  add_lines(y = ~uempmed) %>% 
  add_lines(y = ~fitted(loess(uempmed ~ as.numeric(date))))

[timelyportfolio]

This will overwrite any previous defined layout. Is this intentional?

data(economics, package = "ggplot2")
# basic time-series plot
p <- plot_ly(economics, x = ~date, y = ~uempmed)
# add a loess smoother
p2 <- add_trace(p, y = ~fitted(loess(uempmed ~ as.numeric(date))))
# add a title
p3 <- layout(p2, title = "Median duration of unemployment (in weeks)")
# change the font
layout(p3, font = list(family = "Courier New, monospace"))

The title is lost.

[cpsievert]

Whoops, no, not intentional, I'll fix it, thanks!

[timelyportfolio]

@cpsievert, ok that makes much more sense. Then, in this case, I think documentation, discussion, and examples shoul reflect this philosophy/change. Where is the best place to first communicate this? in ?plot_ly?

[cpsievert]

yea, working on that as we speak :)