Visualizing the Julia Community with Plots

In [ ]:

    

# # to load the data, you'll need to check out my forked repo:
# Pkg.clone("https://github.com/tbreloff/Julia.jl")
# Pkg.checkout("Julia", "tb_github")

# # and you'll want to install Plots
# Pkg.add("Plots")
# Pkg.checkout("Plots")
# Pkg.add("PyPlot")
# Pkg.add("PlotRecipes")
# Pkg.checkout("PlotRecipes")


    

In [ ]:

    




    

In [ ]:

    

# setup Plots and PlotRecipes (graphs and other cool stuff)
using Plots, PlotRecipes
pyplot(size=(700,500), leg=false, guidefont=font(6), titlefont=font(9))


    

In [ ]:

    

# load in the tables
using DataFrames
stardf = readtable(Pkg.dir("Julia","stargazers.csv"))
contribdf = readtable(Pkg.dir("Julia","contributors.csv"))

# a helper method to extract lists from strings
splitlist(s) = filter(str -> str != "", split(s[6:end-3], ","))


    

In [ ]:

    

# count the stargazers
starcounts = map(gazers -> length(splitlist(gazers)), stardf[:gazers]);
sum(starcounts)


    

In [ ]:

    

# Track the number of stars per repo using a Dict
starmap = Dict()
for (repo,cnt) in zip(stardf[:repo], starcounts)
    starmap[repo] = cnt
end
length(starmap)


    

In [ ]:

    

# This is horribly messy.  You don't need to know the details, just that we're
# going to assign package contributors a "Star Value" score, which is just
# a percentage of the stars on a repo assigned by proportion of number of commits.
userstars = Dict{UTF8String,Float64}()
for i=1:size(contribdf,1)
    # extract the package name
    repo = split(contribdf[i,:url], "/")[end]
    repo[end-2:end] == ".jl" || continue
    repo = repo[1:end-3]
    
    # get a list of users (contributors) and the number of contributions
    users = splitlist(contribdf[i,:users])
    nums = splitlist(contribdf[i,:contribs])
    isempty(nums) && continue
    ncontribs = [parse(Int,c) for c in nums]
    
    # assign the stars to users
    pct = ncontribs ./ sum(ncontribs)
    stars = get(starmap, repo, 0)
    pctstars = stars * pct
    for (i,user) in enumerate(users)
        userval = get(userstars, user, 0.0)
        userstars[user] = userval + pctstars[i]
    end
end

# remove if too small
filter!((k,v) -> v>=1, userstars)


    

In [ ]:

    

# what does this data look like?
vals = collect(values(userstars))
plot(
    histogram(vals),
    histogram(vals, bins=500, xaxis=(:log,(1,Inf))),
    scatter(sort(vals), alpha=0.1, yscale=:log),
    layout = @layout [a b; c{0.7h}]
)


    

In [ ]:

    

# Gadfly and IJulia didn't make it onto the list because of formatting issues in Julia.jl
# Gotta give credit to Daniel and Steven!
userstars["dcjones"] += 450
userstars["stevengj"] += 400


    

In [ ]:

    

# get the list of users and calculated star value
users, starvalues = collect(keys(userstars)), collect(values(userstars))


    

In [ ]:

    

# Looking at the highest values, we'll use series annotations to label datapoints.
# Note how the `text` method takes a string and an arbitrarily ordered list of font attributes.
topidx = sortperm(starvalues, rev=true)[1:50]
anns = [text("$(users[i]): $(round(Int,starvalues[i]))", round(Int, 2+0.4sqrt(starvalues[i]))) for i=topidx]

# We cycle through x values 1-4 with invisible markers.  We use the `yaxis` "magic arg" which takes values in any order.
scatter(1:4, starvalues[topidx], series_annotations=anns, xlims=(0.5,4.5), yaxis =(:log,"Star Value",(NaN,1000)), alpha=0)


    

In [ ]:

    

# grab some table columns
groups, subgroups, repos = stardf[:group], stardf[:subgroup], stardf[:repo];
groupnames = sort(unique(groups))'
N = length(groupnames)

# view a grid of histograms, each group in its own subplot
p = histogram(repos, starcounts, group=groups, layout=N, marker=(6,0.2))


    

In [ ]:

    

# This could look better.  Give titles, remove tick labels, and link all the x and y axes.
# Note: xlink/ylink take a list of subplot indices which a subplot should link axes with.
#       Since it's a vector-type, it applies to ALL the subplots
plot!(title = groupnames, xlims=(0,100), ylink = 1:N, ticks=nothing, bins=4)


    

In [ ]:

    

# Not yet an official feature, filter to only the AI group, and boxplot by subgroups
groupfilter(gname) = filter(i -> groups[i] == gname, 1:length(groups))
boxplot(subgroups, starcounts, markersize = 2, idxfilter=groupfilter("AI"))


    

In [ ]:

    

# Make it easier to see
plot!(ylims=(0,100), xrot=90)


    

In [ ]:

    




    

In [ ]:

    

# cycling inputs, series annotations
# Similar to the contributors above, view the 
n = 50
rng = sortperm(starcounts, rev=true)[1:50]
plot(1:4, starcounts[rng], xlim=(0,5), w=0, series_annotations=repos[rng], yscale=:log2)


    

In [ ]:

    

# PlotText construction
text("Hi",10)


    

In [ ]:

    

# text attributes
idx = findfirst(repos, "Plots")
anns = [repo == "Plots" ? text(repo,40,:red) : text(repo,8,RGBA(0,0,0,0.3)) for repo in repos[rng]]
plot(1:4, starcounts[rng], xlim=(0,5), w=0, yscale=:log2, series_annotations=anns)


    

In [ ]:

    




    

In [ ]:

    

# Lets rewrite this as a series recipe
@recipe function f(::Type{Val{:annotations}}, x, y, z)
    seriestype := :scatter
    yscale := :log2
    markeralpha := 0
    xlims := (0.5,4.5)
    series_annotations := x
    x := 1:4
    ()
end
@shorthands annotations

annotations(repos[rng], starcounts[rng])


    

In [ ]: