Regularized F-Measure Maximization for Feature Selection and Classification

Algorithm 2

𝐿₁ regularized F-measure maximization.

1. Given 𝛾, 𝜆, small numbers 𝜀 and 𝛿, 𝐰^𝑡=𝐰⁰, and set

  t = 0 and Ψ={𝑗∶𝑤_𝑗≠0}.

2. While |𝐰^𝑡+1−𝐰^𝑡|>𝜀

  𝐰^𝑡+1=𝐰^𝑡+𝜇((𝜕𝐿𝐹𝜕𝑤_𝑗)_Ψ−𝜆sign(𝑤_𝑖)_Ψ), where 𝜇 is

 the step-size

  Ψ=Ψ∪{𝑗∉Ψ∶|𝜕𝐿𝐹𝑤_𝑗|>𝜆}

  Ψ=Ψ⧵{𝑗∈Ψ∶|𝑤_𝑗|<𝛿}

3. t = t + 1