add example :semantic

Project.toml

@@ -1,7 +1,7 @@
 name = "WordCloud"
 uuid = "6385f0a0-cb03-45b6-9089-4e0acc74b26b"
 authors = ["guoyongzhi <[email protected]>"]
-version = "0.7.2"
+version = "0.7.3"
 
 [deps]
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"

README.md

@@ -49,10 +49,9 @@ paint(wc, "alice.png", ratio=0.5, background=outline(wc.mask, color="purple", li
 ## Recolor
 [![recolor](res/recolor.png)](./examples/recolor.jl)  
 *Run the command `runexample(:recolor)` or `showexample(:recolor)` to get the result.* 
-## Comparison
-[![compare](res/compare.png)](./examples/compare.jl)  
-*Run the command `runexample(:compare)` or `showexample(:compare)` to get the result.* 
-
+## Semantic
+[![semantic](res/semantic.png)](./examples/semantic.jl)  
+*Run the command `runexample(:semantic)` or `showexample(:semantic)` to get the result.* 
 *The variable `WordCloud.examples` holds all available examples.* 
 You can also [**see more examples**](https://github.com/guo-yong-zhi/WordCloud-Gallery) or [**try it online**](https://mybinder.org/v2/gh/guo-yong-zhi/WordCloud.jl/master?filepath=examples.ipynb).  
 # Algorithm Description

examples/embedding.jl → examples/semantic.jl

@@ -1,10 +1,10 @@
-#md# The positions of words can be initialized with pre-trained word vectors so that similar words will appear near each other.
 #md# ### Words
 using WordCloud
 stwords = ["us"];
 words_weights = processtext(open(pkgdir(WordCloud)*"/res/Barack Obama's First Inaugural Address.txt"), stopwords=WordCloud.stopwords_en ∪ stwords)
 words_weights = Dict(zip(words_weights...))
-#md# ### Embeddings
+#md# ### Embedding
+#md# The positions of words can be initialized with pre-trained word vectors so that similar words will appear near each other.
 using Embeddings
 using TSne
 const embtable = load_embeddings(GloVe{:en})
@@ -25,28 +25,51 @@ for k in keys(words_weights)
         println("remove ", k)
     end
 end
-embedded = tsne(hcat(values(wordvec)...)', 2)
-#md# ### WordCloud
+words = keys(wordvec) |> collect
+vectors = hcat(values(wordvec)...)
+embedded = tsne(vectors', 2)
+#md# 
 wc = wordcloud(
     words_weights,
     maskshape = box,
     masksize = (1000, 1000, 0),
-    density=0.3,
-    run = initwords!
+    density = 0.3,
+    colors = 0.3,
+    backgroundcolor = :maskcolor,
+    run = initwords!,
+    # angles = (0,45), font = "Helvetica thin", maskcolor=0.98,
 )
 
 pos = embedded
 mean = sum(pos, dims=1) / size(pos, 1)
 r = maximum(sqrt.(pos[:,1].^2 + pos[:,2].^2 ))
 pos = (pos .- mean) ./ 2r
-sz = collect(size(wc.mask))'
-pos = round.(Int, pos .* sz .+ sz ./ 2)
+sz = collect(reverse(size(wc.mask)))'
+sz0 = collect(getparameter(wc, :masksize)[1:2])'
+pos = round.(Int, pos .* sz0 .+ sz ./ 2)
 
-setpositions!(wc, keys(wordvec)|>collect, eachrow(pos), type=setcenter!)
+setpositions!(wc, words, eachrow(pos), type=setcenter!)
 setstate!(wc, :placewords!)
 generate!(wc, teleporting=false)
-println("results are saved to embedding.png")
-paint(wc, "embedding.png")
+paint(wc, "semantic_embedding.png")
+#md# ![](semantic_embedding.png)  
+#md# ### Clustering
+#md# Words can be further colored according to semantic clustering
+using Clustering
+V = vectors
+G = V' * V
+H = sum(V .^ 2, dims=1)
+D = max.(0, (H .+ H' .- 2G))
+D ./= sum(D)/length(D)
+D .= .√D #the distance matrix
+tree = hclust(D, linkage=:ward)
+lb = cutree(tree, h=2, k=10)
+println("$(length(lb)) words are divided into $(length(unique(lb))) groups")
+#md# 
+colors = parsecolor(:seaborn_dark)
+setcolors!(wc, words, colors[lb.%length(colors).+1])
+recolor!(wc, style=:reset)
+paint(wc, "semantic_clustering.png")
+#md# ![](semantic_clustering.png)  
 wc
-#eval# runexample(:embedding)
-#md# ![](embedding.png)  
+#eval# runexample(:semantic)

src/rendering.jl

@@ -14,6 +14,7 @@ parsecolor(c) = parse(Colorant, c)
 parsecolor(tp::Tuple) = ARGB(tp...)
 parsecolor(gray::Real) = Gray(gray)
 parsecolor(sc::Symbol) = parsecolor.(colorschemes[sc].colors)
+parsecolor(sc::AbstractArray) = parsecolor.(sc)
 
 issvg(d) = d isa Drawing && d.surfacetype==:svg
 const SVGImageType = Drawing