Data gets dropped in add_trace and consequently points have the wrong color if NA is present

[Noskario]

Have a look at the following code:

library(plotly)
x = c(NA, 2, 3, 4, 5)
color = c(0, 0, 0.5, 1, 1)
# If you try to plot it this way, you get mismatched colors. The NA seems to have been silently dropped and then the wrong indices aligned. The colors of points 2 and 3 are now blue, of 4 is purple and of 5 is green.
plotly::plot_ly()|>
    plotly::add_trace(
      type = "scatter",
      mode = "markers",
      x = x,
      y = 0,
      marker = list(
        color = color,
        cmin = 0,
        cmax = 1,
        colorscale = list(list(0, "blue"), list(0.5, "purple"), list(1, "green"))
      )
    ) |>
    plotly::layout(
      showlegend = FALSE
    ) |>
    plotly::hide_colorbar() 

It produces a plot with wrong colors:

I think this is unacceptable behavior, especially since this is not what happens if you do it in the what I consider the classical way:

data.frame(x=x, color=color) |> 
  plot_ly(x=~x, color=~color, y=0, colors = c("blue", "purple", "green")) |> 
  hide_colorbar()

[ismirsehregal]

"silently" is not accurate. I'm getting a warning running your code:

Warning message:
Ignoring 1 observations 

Accordingly, I'd suggest to clean up the input data or pass a data.frame directly to add_trace:

library(plotly)
x = c(NA, 2, 3, 4, 5)
color = c(0, 0, 0.5, 1, 1)

plotly::plot_ly()|>
  plotly::add_trace(data = data.frame(x=x, color=color), x=~x, color=~color, y=0, colors = c("blue", "purple", "green")) |> # inherit = FALSE depending on your usecase
  hide_colorbar()

[Noskario]

Thank you for the proposition to use a dataframe in add_trace.

Sorry for not seeing the warning message. The reason I did not see it, is that I wrapped the plot into a function that styles the plot with our company-design, like this:

library(plotly)
x = c(NA, 2, 3, 4, 5)
color = c(0, 0, 0.5, 1, 1)
stylePlot <- function(p) {
  # among other things, style the colorbar
  # if there is no colorbar, I get a warning.
  # The intent of the suppressWarnings was to get rid of this single warning only
  # However, strangely I do not get any warning at all, if I use this function
  suppressWarnings({
    p <- plotly::colorbar(p, tickfont = list(color = "pink"), 
                          titlefont = list(color = "red"), 
                          tickcolor = "red")
  })
  p
}

plot_ly()|>
plotly::add_trace(
    type = "scatter",
    mode = "markers",
    x = x,
    y = 0,
    marker = list(
      color = color,
      cmin = 0,
      cmax = 1,
      colorscale = list(list(0, "blue"), list(0.5, "purple"), list(1, "green"))
    )
  ) |> 
  stylePlot() |> 
  plotly::layout(
    showlegend = FALSE
  ) |>
  plotly::hide_colorbar() 

I wrongly assumed that the suppressWarnings only suppresses what is inside it, but it seems to also affect the code outside of it.

Please feel free to close this issue if you feel this is not a problem. I personally do not like the behavior of wrongly aligning the data.

[ismirsehregal]

I personally do not like the behavior of wrongly aligning the data.

You are passing two separate vectors to the add_trace call, accordingly the function "does not know" about any alignment (in contrast to passing a data.frame or a named vector representing the mapping).

I think your example is a little confusing. On the one hand, you are using a colorscale indicating a continuous color gradient, on the other hand you are aming at displaying the x values 4 and 5 using the same color which suggests a discrete assignment.

If you are looking for a solution to represent categorical data I'd suggest passing a named vector to the colors parameter, as done here or here, instead of providing a colorscale and hiding it subsequently:

library(plotly)

value = c(NA, 2, 3, 4, 5)
variable = paste0("var_", c(0, 0, 1, 2, 2))
DF <- data.frame(value, variable)
color_mapping <- setNames(c("blue", "purple", "green"), sort(unique(variable)))

plotly::plot_ly() |> plotly::add_trace(
  data = DF,
  x =  ~ value,
  color =  ~ variable,
  y = 0,
  colors = color_mapping
  # , inherit = FALSE # depending on your usecase
) |> layout(showlegend = FALSE)

PS: I don't have the required permissions to close this issue. I'm just another user.