Fix Extrapolation (#20)

* nan out geometry * combine all approaches and fixes * try to let conda magically install matplotlib * don't let PythonCall get in the way! * clean up extrapolation * unique names * install mne if missing * add PyCall * docs + cleanup * fix docs * clean up plots * clean up plots * don't use Conda + add compat for scatteredinterpolations * delete cache for now * add back cache * Maybe that'll reset the cache? * Update src/core-recipe.jl Co-authored-by: Phillip Alday <[email protected]> * use multiple dispatch * remove test.png * PyMNE build tweak * missed a step * try again * another attempt * whyyy * okay sure * percy tweak * more percy tweaks Co-authored-by: Phillip Alday <[email protected]>
MakieOrg · Aug 30, 2022 · b32c1b5 · b32c1b5
1 parent f525872
commit b32c1b5
Show file tree

Hide file tree

Showing 12 changed files with 387 additions and 104 deletions.
diff --git a/.github/workflows/CI.yml b/.github/workflows/CI.yml
@@ -5,6 +5,7 @@ on:
       - master
     tags: '*'
   pull_request:
+
 concurrency:
   # Skip intermediate builds: always.
   # Cancel intermediate builds: only if it is a pull request build.
@@ -30,10 +31,12 @@ jobs:
           version: ${{ matrix.version }}
           arch: ${{ matrix.arch }}
       - uses: julia-actions/cache@v1
+        with:
+          cache-registries: "true"
       - uses: julia-actions/julia-buildpkg@v1
       - uses: julia-actions/julia-runtest@v1
       - name: Percy Upload
-        run: "npx percy upload ./test/test_images"
+        run: 'npx @percy/cli upload ./test/test_images'
         env:
           PERCY_TOKEN: ${{ secrets.PERCY_TOKEN }}
       - uses: julia-actions/julia-processcoverage@v1
@@ -53,4 +56,3 @@ jobs:
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           DOCUMENTER_KEY: ${{ secrets.DOCUMENTER_KEY }}
-
diff --git a/Project.toml b/Project.toml
@@ -12,19 +12,20 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Makie = "ee78f7c6-11fb-53f2-987a-cfe4a2b5a57a"
 Parameters = "d96e819e-fc66-5662-9728-84c9c7592b0a"
 PyCall = "438e738f-606a-5dbb-bf0a-cddfbfd45ab0"
+ScatteredInterpolation = "3f865c0f-6dca-5f4d-999b-29fe1e7e3c92"
 SciPy = "ebc72ef8-9537-4fb0-b64e-ac76025fed2d"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 
 [compat]
-julia = "1"
 Delaunay = "1.2"
 Dierckx = "0.5"
 GeometryBasics = "0.4"
 Makie = "0.17.8"
 Parameters = "0.12"
 PyCall = "1.93"
+ScatteredInterpolation = "0.3.6"
 SciPy = "0.1"
-
+julia = "1"
 
 [extras]
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"

diff --git a/docs/src/functions.md b/docs/src/functions.md
@@ -4,7 +4,3 @@
 TopoPlots.enclosing_geometry
 TopoPlots.labels2positions
 ```
-
-```@docs
-TopoPlots.pad_boundary!
-```
diff --git a/docs/src/general.md b/docs/src/general.md
@@ -7,8 +7,7 @@ At the core of TopoPlots.jl is the `topoplot` recipe, which takes an array of me
 TopoPlots.topoplot
 ```
 
-
-## Interpolators
+## Interpolation
 
 The recipe supports different interpolation methods, namely:
 
@@ -32,21 +31,56 @@ using TopoPlots, CairoMakie
 data, positions = TopoPlots.example_data()
 
 f = Figure(resolution=(1000, 1000))
-interpolators = [DelaunayMesh(), ClaughTochter(), SplineInterpolator()]
+
+interpolators = [
+    DelaunayMesh() ClaughTochter();
+    SplineInterpolator() NullInterpolator()]
+
 data_slice = data[:, 360, 1]
 
-for (i, interpolation) in enumerate(interpolators)
-    j = i == 3 ? (:) : i
+for idx in CartesianIndices(interpolators)
+    interpolation = interpolators[idx]
     TopoPlots.topoplot(
-        f[((i - 1) ÷ 2) + 1, j], data_slice, positions;
+        f[Tuple(idx)...], data_slice, positions;
         contours=true,
         interpolation=interpolation,
         labels = string.(1:length(positions)), colorrange=(-1, 1),
+        label_scatter=(markersize=10,),
         axis=(type=Axis, title="$(typeof(interpolation))()",aspect=DataAspect(),))
 end
 f
 ```
 
+## Extrapolation
+
+There are currently just two extrapolations: None (`NullExtrapolation()`) and a geometry based one:
+
+```@docs
+TopoPlots.GeomExtrapolation
+```
+
+The extrapolations in action:
+
+```@example 1
+data, positions = TopoPlots.example_data()
+titles = ["No Extrapolation", "Rect", "Circle"]
+data_slice = data[:, 340, 1]
+f = Figure(resolution=(900, 300))
+for (i, extra) in enumerate([NullExtrapolation(), GeomExtrapolation(enlarge=3.0), GeomExtrapolation(enlarge=3.0, geometry=Circle)])
+    pos_extra, data_extra, rect_extended, rect = extra(positions, data_slice)
+    geom = extra isa NullExtrapolation ? Rect : extra.geometry
+    # Note, that enlarge doesn't match (the default), the additional points won't be seen and masked by `bounding_geometry` and `enlarge`.
+    enlarge = extra isa NullExtrapolation ? 1.0 : extra.enlarge
+    ax, p = topoplot(f[1, i], data_slice, positions; extrapolation=extra, bounding_geometry=geom, enlarge=enlarge, axis=(aspect=DataAspect(), title=titles[i]))
+    scatter!(ax, pos_extra, color=data_extra, markersize=10, strokewidth=0.5, strokecolor=:white, colormap = p.colormap, colorrange = p.colorrange)
+    lines!(ax, rect_extended, color=:black, linewidth=4)
+    lines!(ax, rect, color=:red, linewidth=1)
+end
+resize_to_layout!(f)
+f
+```
+
+
 ## Interactive exploration
 
 `DelaunayMesh` is best suited for interactive data exploration, which can be done quite easily with Makie's native UI and observable framework:

diff --git a/src/TopoPlots.jl b/src/TopoPlots.jl
@@ -2,14 +2,15 @@ module TopoPlots
 
 using Makie
 using SciPy
-using Delaunay
-using Dierckx
 using LinearAlgebra
 using Statistics
 using GeometryBasics
 using GeometryBasics: origin, radius
 using Parameters
 using InteractiveUtils
+using Delaunay
+using Dierckx
+using ScatteredInterpolation
 
 assetpath(files...) = normpath(joinpath(dirname(@__DIR__), "assets", files...))
 
@@ -24,10 +25,13 @@ end
 
 # Write your package code here.
 include("interpolators.jl")
+include("extrapolation.jl")
 include("core-recipe.jl")
 include("eeg.jl")
 
 # Interpolators
-export ClaughTochter, SplineInterpolator, DelaunayMesh
+export ClaughTochter, SplineInterpolator, DelaunayMesh, NullInterpolator
+# Extrapolators
+export GeomExtrapolation, NullExtrapolation
 
 end
diff --git a/src/core-recipe.jl b/src/core-recipe.jl
@@ -4,9 +4,10 @@
         colorrange = Makie.automatic,
         sensors = true,
         interpolation = ClaughTochter(),
+        extrapolation = GeomExtrapolation(),
         bounding_geometry = Circle,
+        enlarge = 1.2,
         markersize = 5,
-        padding = 0.1,
         pad_value = 0.0,
         resolution = (512, 512),
         labels = nothing,
@@ -27,17 +28,18 @@ Creates an irregular interpolation for each `data[i]` point at `positions[i]`.
 * `colorrange = automatic`
 * `labels::Vector{<:String}` = nothing: names for each data point
 * `interpolation::Interpolator = ClaughTochter()`: Applicable interpolators are $(join(subtypes(TopoPlots.Interpolator), ", "))
-* `bounding_geometry = Circle`: the geometry added to the points, to create a smooth boundary. Can be `Rect` or `Circle`.
-* `markersize = 5`: size of the points defined by positions
-* `padding = 0.1`: padding applied to `bounding_geometry`
-* `pad_value = 0.0`: data value filled in for each added position from `bounding_geometry`
+* `extrapolation = GeomExtrapolation()`: Extrapolation method for adding additional points to get less border artifacts
+* `bounding_geometry = Circle`: A geometry that defines what to mask and the x/y extend of the interpolation. E.g. `Rect(0, 0, 100, 200)`, will create a `heatmap(0..100, 0..200, ...)`. By default, a circle enclosing the `positions` points will be used.
+* `enlarge` = 1.2`, enlarges the area that is being drawn. E.g., if `bounding_geometry` is `Circle`, a circle will be fitted to the points and the interpolation area that gets drawn will be 1.2x that bounding circle.
 * `resolution = (512, 512)`: resolution of the interpolation
 * `label_text = false`:
     * true: add text plot for each position from `labels`
     * NamedTuple: Attributes get passed to the Makie.text! call.
 * `label_scatter = false`:
     * true: add point for each position with default attributes
     * NamedTuple: Attributes get passed to the Makie.scatter! call.
+* `markersize = 5`: size of the points defined by positions, shortcut for label_scatter=(markersize=5,)
+
 * `contours = false`:
     * true: add scatter point for each position
     * NamedTuple: Attributes get passed to the Makie.contour! call.
@@ -54,6 +56,7 @@ topoplot
 # Handle the nothing/bool/attribute situation for e.g. contours/label_scatter
 plot_or_defaults(value::Bool, defaults, name) = value ? defaults : nothing
 plot_or_defaults(value::Attributes, defaults, name) = merge(value, defaults)
+
 function plot_or_defaults(value, defaults, name)
     error("Attribute $(name) has the wrong type: $(typeof(value)).
           Use either a bool to enable/disable plotting with default attributes,
@@ -65,48 +68,65 @@ macro plot_or_defaults(var, defaults)
 end
 
 function Makie.plot!(p::TopoPlot)
-    npositions = Observable(0; ignore_equal_values=true)
-    geometry = lift(enclosing_geometry, p.bounding_geometry, p.positions, p.padding; ignore_equal_values=true)
-    p.geometry = geometry # store geometry in plot object, so others can access it
+    Obs(x) = Observable(x; ignore_equal_values=true) # we almost never want to trigger updates if value stay the same
+    npositions = Obs(0)
+
     # positions changes with with data together since it gets into convert_arguments
     positions = lift(identity, p.positions; ignore_equal_values=true)
-    padded_position = lift(positions, geometry, p.resolution; ignore_equal_values=true) do positions, geometry, resolution
-        points_padded = append!(copy(positions), decompose(Point2f, geometry))
-        npositions[] = length(points_padded)
-        return points_padded
-    end
+    geometry = lift(enclosing_geometry, p.bounding_geometry, positions, p.enlarge; ignore_equal_values=true)
 
-    xg = Observable(LinRange(0f0, 1f0, p.resolution[][1]); ignore_equal_values=true)
-    yg = Observable(LinRange(0f0, 1f0, p.resolution[][2]); ignore_equal_values=true)
+    xg = Obs(LinRange(0f0, 1f0, p.resolution[][1]))
+    yg = Obs(LinRange(0f0, 1f0, p.resolution[][2]))
 
-    f = onany(geometry, p.resolution) do geom, resolution
-        xmin, ymin = minimum(geom)
-        xmax, ymax = maximum(geom)
+    f = onany(geometry, p.resolution) do geometry, resolution
+        (xmin, ymin), (xmax, ymax) = extrema(geometry)
         xg[] = LinRange(xmin, xmax, resolution[1])
         yg[] = LinRange(ymin, ymax, resolution[2])
         return
     end
+
     notify(p.resolution) # trigger above (we really need `update=true` for onany)
 
-    padded_data = lift(pad_data, p.data, npositions, p.pad_value)
+    p.geometry = geometry # store geometry in plot object, so others can access it
+
+    padded_pos_data_bb = lift(p.extrapolation, p.positions, p.data) do extrapolation, positions, data
+        return extrapolation(positions, data)
+    end
+
+    colorrange = lift(p.data, p.colorrange) do data, crange
+        if crange isa Makie.Automatic
+            return Makie.extrema_nan(data)
+        else
+            return crange
+        end
+    end
 
     if p.interpolation[] isa DelaunayMesh
         # TODO, delaunay works very differently from the other interpolators, so we can't switch interactively between them
-        m = lift(delaunay_mesh, padded_position)
-        mesh!(p, m, color=padded_data, colorrange=p.colorrange, colormap=p.colormap, shading=false)
+        m = lift(delaunay_mesh, p.positions)
+        mesh!(p, m, color=p.data, colorrange=colorrange, colormap=p.colormap, shading=false)
     else
-        data = lift(p.interpolation, xg, yg, padded_position, padded_data) do interpolation, xg, yg, points, data
-            return interpolation(xg, yg, points, data)
+        data = lift(p.interpolation, xg, yg, padded_pos_data_bb, geometry) do interpolation, xg, yg, (points, data, _, _), geometry
+            z = interpolation(xg, yg, points, data)
+            for xy_idx in CartesianIndices(z)
+                xi, yi = Tuple(xy_idx)
+                xy = Point2f(xg[xi], yg[yi])
+                if !(xy in geometry)
+                    z[xy_idx] = NaN
+                end
+            end
+            return z
         end
-        heatmap!(p, xg, yg, data, colormap=p.colormap, colorrange=p.colorrange, interpolate=true)
+
+        heatmap!(p, xg, yg, data, colormap=p.colormap, colorrange=colorrange, interpolate=true)
         contours = to_value(p.contours)
         attributes = @plot_or_defaults contours Attributes(color=(:black, 0.5), linestyle=:dot, levels=6)
-        if !isnothing(attributes)
+        if !isnothing(attributes) && !(p.interpolation[] isa NullInterpolator)
             contour!(p, xg, yg, data; attributes...)
         end
     end
     label_scatter = to_value(p.label_scatter)
-    attributes = @plot_or_defaults label_scatter Attributes(markersize=p.markersize, color=p.data, colormap=p.colormap, colorrange=p.colorrange, strokecolor=:black, strokewidth=1)
+    attributes = @plot_or_defaults label_scatter Attributes(markersize=p.markersize, color=p.data, colormap=p.colormap, colorrange=colorrange, strokecolor=:black, strokewidth=1)
     if !isnothing(attributes)
         scatter!(p, p.positions; attributes...)
     end
@@ -119,49 +139,3 @@ function Makie.plot!(p::TopoPlot)
     end
     return
 end
-
-"""
-    enclosing_geometry(G::Type{<: Geometry}, positions, enlarge=0.0)
-
-Returns the Geometry of Type `G`, that best fits all positions.
-The Geometry can be enlarged by 1.x, so e.g. `enclosing_geometry(Circle, positions, 0.1)` will return a Circle that encloses all positions with a padding of 10%.
-"""
-function enclosing_geometry(::Type{Circle}, positions, enlarge=0.0)
-    middle = mean(positions)
-    radius, idx = findmax(x-> norm(x .- middle), positions)
-    return Circle(middle, radius * (1 + enlarge))
-end
-
-function enclosing_geometry(::Type{Rect}, positions, enlarge=0.0)
-    rect = Rect2f(positions)
-    w = widths(rect)
-    padded_w = w .* (1 + 2enlarge)
-    mini = minimum(rect) .- ((padded_w .- w) ./ 2)
-    return Rect2f(mini, padded_w)
-end
-
-"""
-    pad_boundary(::Type{Geometry}, positions, enlarge=0.2) where Geometry
-
-Adds new points to positions, adding the boundary from enclosing all positions with `Geometry`.
-See [`TopoPlots.enclosing_geometry`](@ref) for more details about the boundary.
-"""
-function pad_boundary!(::Type{Geometry}, positions, enlarge=0.2) where Geometry
-    c = enclosing_geometry(Geometry, positions, enlarge)
-    return append!(positions, decompose(Point2f, c))
-end
-
-function pad_data(data::AbstractVector, positions::AbstractVector, value::Number)
-    pad_data(data, length(positions), value)
-end
-
-function pad_data(data::AbstractVector, npositions::Integer, value::Number)
-    ndata = length(data)
-    if npositions == ndata
-        return data
-    elseif npositions < ndata
-        error("To pad the data for new positions, we need more positions than data points")
-    else
-        vcat(data, fill(value, npositions - ndata))
-    end
-end