several classfication model comparison

1. load package

Code

    import MLJ:predict,predict_mode
    using  MLJ,GLMakie,DataFrames,Random
    Random.seed!(1222)

TaskLocalRNG()

2. make data

Code

    function circle_data()
    X, y = make_circles(400; noise=0.1, factor=0.3)
    df = DataFrame(X)
    df.y = y
    return df
    end
    function moons_data()
        X, y = make_moons(400; noise=0.1)
        df = DataFrame(X)
        df.y = y
        return df
    end
    function blob_data()
        X, y = make_blobs(400, 2; centers=2, cluster_std=[1.0, 2.0])
        df = DataFrame(X)
        df.y = y
        return df
    end
    #cat=df1.y|>levels|>unique
    colors=[:green, :purple]

2-element Vector{Symbol}:
 :green
 :purple

3. define function

Code

function plot_origin_data(df)
    fig=Figure()
    ax=Axis(fig[1,1])
    local cat=df.y|>levels|>unique
    
    local colors=[:green, :purple]
    for (i,c) in enumerate(cat)
        d=df[y.==c,:]
        scatter!(ax, d[:,1],d[:,2],color=(colors[i],0.6))
        #@show d
    end
    fig
end

nums=100
function boundary_data(df,;n=nums)
    n1=n2=n
    xlow,xhigh=extrema(df[:,:x1])
    ylow,yhigh=extrema(df[:,:x2])
    tx = LinRange(xlow,xhigh,n1)
    ty = LinRange(ylow,yhigh,n2)
    x_test = mapreduce(collect, hcat, Iterators.product(tx, ty));
    x_test=MLJ.table(x_test')
    return tx,ty,x_test
end

function plot_desc_boudary(fig,ytest,i;df=df1,row=1)
        tx,ty,xs,ys, xtest=boundary_data(df)
        local ax=Axis(fig[row,i],title="$(names[i])")

        contourf!(ax, tx,ty,ytest,levels=length(cat),colormap=:phase)

        for (i,c) in enumerate(cat)
            d=df[y.==c,:]
            scatter!(ax, d[:,1],d[:,2],color=(colors[i],0.6))
        end
        hidedecorations!(ax)
end

plot_desc_boudary (generic function with 1 method)

4. define machine learning models

Code

    using CatBoost.MLJCatBoostInterface
    SVC = @load SVC pkg=LIBSVM   
    KNNClassifier = @load KNNClassifier pkg=NearestNeighborModels
    DecisionTreeClassifier = @load DecisionTreeClassifier pkg=DecisionTree
    RandomForestClassifier = @load RandomForestClassifier pkg=DecisionTree
    CatBoostClassifier = @load CatBoostClassifier pkg=CatBoost
    BayesianLDA = @load BayesianLDA pkg=MultivariateStats
    Booster = @load AdaBoostStumpClassifier pkg=DecisionTree
    
    models=[KNNClassifier,DecisionTreeClassifier,RandomForestClassifier,CatBoostClassifier,BayesianLDA,SVC]
    names=["KNN","DecisionTree","RandomForest","CatBoost","BayesianLDA","SVC"]
   function _fit(df::DataFrame,m)
    X,y=df[:,1:2],df[:,3]
    _,_,xtest=boundary_data(df;n=nums)
    local predict= m==MLJLIBSVMInterface.SVC  ? MLJ.predict : MLJ.predict_mode 
    model=m()
   mach = machine(model, X, y)|>fit!
   yhat=predict(mach, xtest)
   ytest=yhat|>Array|>d->reshape(d,nums,nums)
   return  ytest
end



function plot_desc_boudary(fig,ytest,i;df=df1,row=1)
    tx,ty,_=boundary_data(df)
    local y=df.y
    local ax=Axis(fig[row,i],title="$(names[i])")
    cat=y|>levels|>unique
    contourf!(ax, tx,ty,ytest,levels=length(cat),colormap=:redsblues)

    for (i,c) in enumerate(cat)
        d=df[y.==c,:]
        scatter!(ax, d[:,1],d[:,2],color=(colors[i],0.6))
    end
    hidedecorations!(ax)
    

end

function plot_comparsion(testdata,df;row=1)
    
    for (i,data) in enumerate(testdata)
        plot_desc_boudary(fig,data,i;df=df,row=row)
    end
    fig
end

[ Info: For silent loading, specify `verbosity=0`. 
[ Info: For silent loading, specify `verbosity=0`. 
[ Info: For silent loading, specify `verbosity=0`. 
[ Info: For silent loading, specify `verbosity=0`. 
[ Info: For silent loading, specify `verbosity=0`. 
[ Info: For silent loading, specify `verbosity=0`. 
[ Info: For silent loading, specify `verbosity=0`. 

import MLJLIBSVMInterface ✔
import NearestNeighborModels ✔
import MLJDecisionTreeInterface ✔
import MLJDecisionTreeInterface ✔
import CatBoost ✔
import MLJMultivariateStatsInterface ✔
import MLJDecisionTreeInterface ✔

plot_comparsion (generic function with 1 method)

Code

fig=Figure(resolution=(2100,1000))
function plot_comparsion(testdata,df,row=1)
    
    for i in eachindex(testdata)
        plot_desc_boudary(fig,testdata[i],i;df=df,row=row)
    end
    fig
end



df1=circle_data()

ytest1=[_fit(df1,m) for (i,m) in enumerate(models)]

df2=moons_data()
ytest2=[_fit(df2,m) for (i,m) in enumerate(models)]

df3=blob_data()
ytest3=[_fit(df3,m) for (i,m) in enumerate(models)]

dfs=[df2,df1,df3]
ytests=[ytest2,ytest1,ytest3]

fig=Figure(resolution=(2100,1000))

for (df, data,i)  in zip(dfs,ytests,[1,2,3])
    plot_comparsion(data,df;row=i)
end

fig

[ Info: Training machine(KNNClassifier(K = 5, …), …).
[ Info: Training machine(DecisionTreeClassifier(max_depth = -1, …), …).
[ Info: Training machine(RandomForestClassifier(max_depth = -1, …), …).
[ Info: Training machine(CatBoostClassifier(iterations = 1000, …), …).
[ Info: Training machine(BayesianLDA(method = gevd, …), …).
[ Info: Training machine(SVC(kernel = RadialBasis, …), …).
[ Info: Training machine(KNNClassifier(K = 5, …), …).
[ Info: Training machine(DecisionTreeClassifier(max_depth = -1, …), …).
[ Info: Training machine(RandomForestClassifier(max_depth = -1, …), …).
[ Info: Training machine(CatBoostClassifier(iterations = 1000, …), …).
[ Info: Training machine(BayesianLDA(method = gevd, …), …).
[ Info: Training machine(SVC(kernel = RadialBasis, …), …).
[ Info: Training machine(KNNClassifier(K = 5, …), …).
[ Info: Training machine(DecisionTreeClassifier(max_depth = -1, …), …).
[ Info: Training machine(RandomForestClassifier(max_depth = -1, …), …).
[ Info: Training machine(CatBoostClassifier(iterations = 1000, …), …).
[ Info: Training machine(BayesianLDA(method = gevd, …), …).
[ Info: Training machine(SVC(kernel = RadialBasis, …), …).