R: plot 3d interaction model and observations using plotly - r

Question

I start from a simple interaction model z = 0.5*x*y. The goal is to plot simulated observations with add_markers() as well as the model with add_surface() using the plotly library.
So far I have manage to simulate the data and plot it:
library(plotly)
library(magrittr)
x <- rnorm(10000)
y <- rnorm(10000)
z <- 0.5 * x * y
plot_ly() %>% add_markers(x = x, y = y, z = z, marker = list(size = 1))
This is what the plot looks like:
I then tried to surface plot the model:
plot_ly() %>%
add_markers(x = x, y = y, z = z, marker = list(size = 1)) %>%
add_surface(z = matrix(z, 100, 100))
But it produces something completely different to what it would be expected. This is approximately the result that would be expect from the surface plot:
What am I doing wrong?

Answer 1

An easy workaround is to use:
plot_ly(x = x, y = y, z= z, type = 'mesh3d')
Or using surface plot:
library(tidyverse)
grid <- expand.grid(x = seq(from = -3, to = 3, length.out = 100),
y = seq(from = -3, to = 3, length.out = 100))
grid %>%
mutate(z = 0.5 * x * y) %>%
spread(key = y, value = z) %>%
as.matrix() -> z
plot_ly() %>%
add_surface(z = z, x = unique(grid$x), y = unique(grid$y)) %>%
layout(scene = list(xaxis = list(range = c(-2.5, 2.5)),
yaxis = list(range = c(-2.5, 2.5)))) #odd artifact if all values are plotted - check without layout call
or:
model = lm(z ~ x * y)
grid <- expand.grid(x = seq(from = -3, to = 3, length.out = 100),
y = seq(from = -3, to = 3, length.out = 100))
grid$z = predict(model, grid)
and proceed as above
grid %>%
spread(key = y, value = z) %>%
as.matrix() -> z