add_trace inherits properties in this example

[chriddyp]

zcol = c(1, 3, 2)
textcol = c('A', 'B', 'C')
countrycol = c('Canada', 'Egypt', 'USA')
locationscol = c('CAN', 'EGY', 'USA')
df = data.frame(z=zcol, text=textcol, country=countrycol, locations=locationscol)

p <- plot_ly(df, z = z, text = text, locations = locations, type = 'choropleth') %>%
  add_trace(type="scattergeo", mode="text") 
fig <- p %>% plotly_build

the second trace should only have "type" and "mode" keys, but instead:

> names(fig$data[[2]])
[1] "type"       "inherit"    "z"          "text"       "locations"  "colorbar"   "colorscale"

[cpsievert]

If you don't want add_trace() to inherit properties from plot_ly(), you have to specify inherit = FALSE

zcol = c(1, 3, 2)
textcol = c('A', 'B', 'C')
countrycol = c('Canada', 'Egypt', 'USA')
locationscol = c('CAN', 'EGY', 'USA')
df = data.frame(z=zcol, text=textcol, country=countrycol, locations=locationscol)

p <- plot_ly(df, z = z, text = text, locations = locations, type = 'choropleth', inherit = FALSE) %>%
    add_trace(type="scattergeo", mode="text") 
fig <- p %>% plotly_build
fig$data[[2]]
#> $type
#> [1] "scattergeo"
#> 
#> $mode
#> [1] "text"

Do you think the default should be inherit = FALSE?

[chriddyp]

ahhh got it. Not sure about the default, but I think FALSE is the least surprising option and explicit option:

df <- data.frame(x1=c(1, 2, 3), y1=c(3, 1, 6), t1=c('a', 'b', 'c'), x2=c(2, 1, 5), y2=c(5, 4, 5), t2=('x', 'y', 'z')

plot_ly(df, x=x1, y=y1, text=t1) %>% add_trace(x=x2, y=y2)

In this case, the user might've forgotten to add text=t2 in the second call, and might be misled by the inherited hover text.

[cpsievert]

I'm starting to agree that FALSE might be a better default. Also, inherit=TRUE should mean to inherit the arguments, not actual values. This has important consequences when we have a different data frame for each traces

library(plotly)
library(dplyr)
p <- iris %>%
      count(Species) %>%
      plot_ly(x = Species, y = n, opacity = 0.5, type = "bar") %>%
      layout(barmode = "overlay", showlegend = FALSE)

s <- count(iris[sample(nrow(iris), 10), ], Species)
add_trace(p, data = s)

versus

add_trace(p, x = Species, y = n, data = s)