inst_ideone_logistic.rb

#----------------------
#ロジスティック回帰
 
require "matrix"
 
def sigmoid(x)
    1.0/(1.0+Math::exp(-x))
end
 
class LearnLogistic
    def initialize(dim,eta)
        @dim=dim
        @eta=eta
        @weight=Vector.elements([0.0]*dim)
    end
    def test(val)
        (@weight.covector*val.covector.t).det
    end
    def update(data,cls)
        yn=sigmoid(test(data))
        @weight=@weight-@eta*(yn-cls)*data
    end
    def learn(data,cls,n)
 
    end
    def dim;@dim;end
    def weight;@weight;end
    def weight=(w);@weight=w;end
    def eta;@eta;end
    def eta=(w);@eta=w;end
end
 
 
#データの入力
data=[]
while cin=gets
    cl=cin.split(",")
    data+=[[cl[0]=="t",[cl[1].to_i,cl[2].to_i,1]]]
end
 
#データ集合から取り出し
size=data.size
dim=3
ax=[]
at=[]
data.each{|e|
    ax+=[e[1]]
    at+=[e[0]==true ? 1.0 : 0.0]
}
mx=Matrix.columns(ax)       # 入力データ
vt=Vector.elements(at)      # クラス分類
 
#-------------------------------
#計算
 
mlearn=LearnLogistic.new(dim,0.2)
 
maxacc=nil
maxw=nil
100.times{|cnt|
    # Learning
    size.times{|i|
        mlearn.update(mx.column(i),vt[i])
    }
    # Test
    acc=0
    size.times{|i|
        t=(mlearn.test(mx.column(i)) > 0) == data[i][0]
        acc+=1 if t
    }
    if acc==size
        puts "solve "+cnt.to_s
        maxacc=nil
        break
    end
    mlearn.eta=mlearn.eta*0.95 if mlearn.eta > 1.0e-3
}
puts "Result Vector" ; p mlearn.weight
 
puts "Test..."
 
#試しに計算してみる
size.times{|i|
    puts sprintf("res=%8.3f Data={%s}",mlearn.test(mx.column(i)),data[i]*",")
}