Bug of Scatter3D plots for only one point with multilevel?

[BoyaoZhang]

plot_ly(x = iris[1,1], y = iris[1,2], z = iris[1,3], type = "scatter3d", symbol = iris[1,5])

[kcf-jackson]

I think the issue is that you supplied a factor to 'symbol'. Using characters string instead would fix the issue.
plot_ly(x = iris[1,1], y = iris[1,2], z = iris[1,3], type = "scatter3d", symbol = as.character(iris[1,5]))

[BoyaoZhang]

In fact I want to perform a machine learning algorithm on the iris data. According to the result, some points may be incorrect predicted. And I want to figure out this kind of points. For example, if the first data is miss-classified, I still want to have the same symbol shape, just want to change the color. Like this,

iris_without1 = iris[-1,]
iris_1 = iris[1,]
p = plot_ly(x = iris_without1[,1], y = iris_without1[,2], z = iris_without1[,3],
            type = "scatter3d", symbol = iris_without1[,5], marker = list(size = 3))
add_trace(p, x = iris_1[,1], y = iris_1[,2], z = iris_1[,3],
          type = "scatter3d", symbol = iris_1[,5], 
          marker = list(size = 3, color = "black"))

You can see the square-shaped black point (miss-classified) in setosa group. This is in fact what I want in the end.

But I don't want to change it by giving a symbol name, I hope it perform automatically.

[kcf-jackson]

If you want to remove the legend of the black dot, simply add "showlegend = FALSE" in the "add_trace" call.

iris_without1 = iris[-1,]
iris_1 = iris[1,]
p = plot_ly(x = iris_without1[,1], y = iris_without1[,2], z = iris_without1[,3],
            type = "scatter3d", symbol = as.character(iris_without1[,5]), marker = list(size = 3))
add_trace(p, x = iris_1[,1], y = iris_1[,2], z = iris_1[,3],
          type = "scatter3d", symbol = as.character(iris_1[,5]), 
          marker = list(size = 5, color = "black"), showlegend = FALSE)

I think your questions are not development related, perhaps you could consider asking them in stackoverflow instead.

[BoyaoZhang]

Sorry I asked here, but I still believe it must be a bug for single point with multi levels.
I think I chose a bad example, if I choose the 70th observation which belongs to versicolor. In the plot it should be a cross, but with as.character function it resolves as circle.

iris_without1 = iris[-70,]
iris_1 = iris[70,]
p = plot_ly(x = iris_without1[,1], y = iris_without1[,2], z = iris_without1[,3],
            type = "scatter3d", symbol = as.character(iris_without1[,5]), 
            marker = list(size = 3))
add_trace(p, x = iris_1[,1], y = iris_1[,2], z = iris_1[,3],
          type = "scatter3d", symbol = as.character(iris_1[,5]), 
          marker = list(size = 5, color = "black"), showlegend = FALSE)

[kcf-jackson]

As far as I see, plotly is doing exactly what you instructed it to do. Here is my understanding.

1. Plotly treats the data in 'add_trace' independently of the data you use in 'plot_ly', hence plotly produces two sets of symbols.
2. In 'add_trace', the data only has 1 point, the default 'circle' symbol is used.

To solve your problem, I thought about using 'as.numeric' instead of 'as.character', but somehow plotly doesn't convert the numbers according to https://plot.ly/r/reference/#box-marker-symbol.
Though a simple fix can be found as follows:

data(iris)
symbol_3set <- c("circle", "cross", "square")

iris_without1 = iris[-70,]
iris_1 = iris[70,]
plot_ly(data = iris_without1, x = Sepal.Length, y = Sepal.Width, z = Petal.Length,
            type = "scatter3d", mode = "markers", color = Species,
            marker = list(size = 3, symbol = symbol_3set[as.numeric(Species)]))
add_trace(data = iris_1, x = Sepal.Length, y = Sepal.Width, z = Petal.Length,
          type = "scatter3d", mode = "markers", 
          marker = list(size = 5, color = "black", symbol = symbol_3set[as.numeric(Species)]), 
          showlegend = FALSE)

[BoyaoZhang]

I must thank you for your patient. It goes very close to the end version which I want.
You know, iris is a very simple example. If I change the data set which has a lot of different levels, do you mean I should always type all the symbol names?
Why plotly couldn't convert the numeric symbol values? In https://plot.ly/r/reference/#box-marker-symbol, the symbols are coded both for character and for numeric, aren't they?
In the following example, we can easily see the legend symbols are correct coded.

test = iris[c(1, 2, 51, 52, 101, 102),]
plot_ly(x = test$Sepal.Length, y = test$Sepal.Width, z = test$Petal.Length,
        type = "scatter3d", mode = "markers", color = test$Species,
        marker = list(symbol = c(1,1,2,2,3,3)))

[kcf-jackson]

One uses colours if there are many levels as plotly only supports up to 8 symbols as far as I know.
And according to the documentation in the source code,

#' @param symbol Either a variable name or a (discrete) vector to use for symbol encoding.
#' @param symbols A character vector of symbol types. Possible values:
#' 'dot', 'cross', 'diamond', 'square', 'triangle-down', 'triangle-left', 'triangle-right', 'triangle-up' 

there is no mention of allowing numeric encoding of symbols. The reference is possibly outdated.

[BoyaoZhang]

I have run the following code to test these symbols, but any code couldn't generate the correct answer.

test = iris
test$Species = sample(1:8, 150, replace = T)
test$Species = as.factor(test$Species)

One can find in the legend labels that there are only 4 symbols

plot_ly(x = test$Sepal.Length, y = test$Sepal.Width, z = test$Petal.Length,
        type = "scatter3d", mode = "markers", symbol = test$Species)

If I put the symbol parameter into the marker list, the symbols in the legend displays correctly, but none of the point are really with the symbol converted.

plot_ly(x = test$Sepal.Length, y = test$Sepal.Width, z = test$Petal.Length,
        type = "scatter3d", mode = "markers", color = test$Species,
        marker = list(symbol = as.numeric(test$Species)))

Now I added the symbolset into the function. The output is almost same with the first plot but the incorrect legend labels.

symbolset = c('dot', 'cross', 'diamond', 'square', 'triangle-down', 'triangle-left', 'triangle-right', 'triangle-up')

plot_ly(x = test$Sepal.Length, y = test$Sepal.Width, z = test$Petal.Length,
        type = "scatter3d", mode = "markers",
        symbol = symbolset[as.numeric(test$Species)])

Finally, the following code convert some points correctly, but some not. You can see, the triangle and dot symbols don't work. If I replace the symbol "dot" with the name "circle", than it works well for "circle". What is wrong with the outrageously symbols?

plot_ly(x = test$Sepal.Length, y = test$Sepal.Width, z = test$Petal.Length,
        type = "scatter3d", mode = "markers", color = test$Species,
        marker = list(symbol = symbolset[as.numeric(test$Species)]))

[cpsievert]

pretty sure this is a duplicate of #577