plot_predict( ) function

[ReneSkukies]

Since we now have a predict() function, it would be nice to have a function that easily plots the predicted data. Maybe overlayed/ subplotted with the actual data. I achieved it in the following way:

using Unfold, CairoMakie

# Predict the data from model m
pred_data_stim = predict(
    m,
    Unfold.formulas(m),
    Unfold.events(m);
    overlap = true,
    keep_basis = [Unfold.basisname(Unfold.formulas(m))[2]]
)

pred_data_resp = predict(
    m,
    Unfold.formulas(m),
    Unfold.events(m);
    overlap = true,
    keep_basis = [Unfold.basisname(Unfold.formulas(m))[1]]
)

fig = Figure()
ax1 = Axis(fig[1], width=Relative(1), height=Relative(0.5), halign=0.0, valign=1)
ax2 = Axis(fig[1], width=Relative(1), height=Relative(0.5), halign=0.0, valign=0.0)
l = (500, 1500) # limits of the plot

# Plot continuous data
lines!(ax1, data, color = :black)

# Plot non-overlapping data
lines!(ax2, pred_data_stim[1,:], color = (:blue, 0.5))
lines!(ax2, pred_data_resp[1,:], color = (:red, 0.5))


# Set limits and hide stuff
axes = [ax1, ax2]
for a in axes; xlims!(a, l...); hidedecorations!(a); hidespines!(a); end

# Stimulus marker
colors = [:blue, :green]
for (i,c) in enumerate(unique(evts.condition))
vlines!(Axis(fig[pos...]),@rsubset(evts, :condition .== c, :event .== 'S').latency, color = (colors[i], 0.5))
xlims!(l...)
hidedecorations!(current_axis())
hidespines!(current_axis())
end

# Response marker
vlines!(Axis(fig[pos...]),@rsubset(evts,:event .== 'R').latency, color = (:red, 0.5))
xlims!(l...)
hidedecorations!(current_axis())
hidespines!(current_axis())

current_figure()

In this case it looks like this for me (but it's specific for my figure, thus no decorations)

This way one could visually inspect how much of the data is explained by the model (when plotting predicted data with overlap), how much overlap is in the data, etc.
Additionally, this could be nice for future figures and presentations.

[behinger]

I like it. One issue I see, which I also see with the DesignMatrix, is that if we have a long dataset, we don't really know where (or "when in time") to plot, and plotting everything is maybe a bit much (even though possible outside CairoMakie+SVG).

Maybe it is time time to implement a datascroll after all (similar to https://github.com/JuliaHealth/NeuroAnalyzer.jl)