adding points to a surface reverses x and y axis

[romunov]

In trying to replicate some 3D marginal plots of a model, I notice that plot_ly swaps axes when adding points.

When using persp, I get a figure which I consider correct. meshout changes with size, but not distance.

but when mapping size and distance to x and y respectively, axes appear to be switched.

When manually switching them, plot appears fine.

Is this something you guys can look into? Below is a fully working example.

library(plotly)
library(ggplot2)
library(reshape2)
library(rgl)

N <- 50
B0 <-  1.2
B1 <-  1.5
B2 <- -0.5

set.seed(357)
count <- rpois(N, 10)

distance <- seq(-0.5, 0.5,,length(count))
size <- (log(count) - B0 - B2 * distance) / B1
xy <- data.frame(count, size, distance)

mdl <- glm(count ~ size * distance, family = poisson(link = log), data = xy)
summary(mdl)

ggplot(xy, aes(x = size, y = count, color = distance)) +
  geom_point()

npts <- 50
s.size <- seq(from = min(xy$size), to = max(xy$size), length.out = npts)
s.dist <- seq(from = min(xy$distance), to = max(xy$distance), length.out = npts)
mesh <- expand.grid(
  size = s.size,
  distance = s.dist
)
mesh$count <- predict(mdl, newdata = mesh, type = "response")
head(mesh)

meshout <- as.matrix(
  dcast(mesh, formula = size ~ distance, value.var = "count")[, -1]
)

pmat <- persp(x = s.size, y = s.dist, z = meshout, # size and distance mapped OK
      theta = 0, phi = 30,
      col = "grey90", border = "grey50")
points(trans3d(x = xy$size, y = xy$dist, z = xy$count, pmat = pmat))

plot3d(x = xy$size, y = xy$distance, z = xy$count) # check, distance and size are plotted correctly
# https://stackoverflow.com/questions/5468280/scale-a-series-between-two-points#comment49916403_5468527
range02 <- function(x, newMin, newMax){ (x - min(x))/(max(x)-min(x)) * (newMax - newMin) + newMin }

xy$scaled_size <- range02(xy$size, newMin = 0, newMax = 50)
xy$scaled_distance <- range02(xy$distance, newMin = 0, newMax = 50)

# size and distance swapped
plot_ly(z = ~ meshout) %>% 
  add_surface(name = "Incorrect mapping of x and y") %>%
  add_markers(data = xy, x = ~scaled_size, y = ~scaled_distance, z = ~count,
              marker = list(size = 5, color = "black"), name = "Incorrect mapping of x and y") 

# when switching mapping of x and y, plots fine
plot_ly(z = ~ meshout) %>% 
  add_surface(name = "Incorrect mapping of x and y") %>%
  add_markers(data = xy, x = ~scaled_distance, y = ~scaled_size, z = ~count,
              marker = list(size = 5, color = "black"), name = "Incorrect mapping of x and y")

[JakeJing]

I recently come across the same problem only in plotly chart studio. The script works perfectly on my own computer, and I can also share it via api_create. But whenever I click "open in editor" in Plotly chart studio. It will switch x and y and make the points fall out of the fitted surface.

I also attached a working example here. I guess it is the problem of plotly chart studio. Right?

[ningyile]

I also have the same problem. Even when the 'add_markers' function is not applied, the corresponding relationship between the x-axis and y-axis in a 3d surface plot is still inverse. So unless you switch parameters of the x and y axis in the function of 'add_surface', the mapping disorder will not be solved. This is flawed for the visualization of trained models, it is disturbing to think that many people do not notice the X and Y mapping bug. I have noticed that this bug has been reported for 5 years (Romanov @romunov reported it in 2017) and I hope anyone who can fix it in the near future. The same data is plotted correctly with rgl::persp3d and persp function.

[AxelCode-R]

i have the same bug.... but only sometimes. Im ploting multiple functions and some have only the axis reversed which is anoying but correct and some have reversed axis but the plot shows the function in the right order, which is definitly wrong!