Table 1.

Graphical lasso algorithm.

1. Initialize W = S + λI. Note that the diagonal of W is fixed in what follows. 2. Repeat the following steps until W is converged.    a. Partition the matrix W such that the target column is last.    b. Solve the lasso problem using the coordinate descent algorithm.    c. Update $w_{12}=W_{11}\text{β}.$ 3. In the final cycle, calculate ${\theta }_{12}=-\beta {\theta }_{22}$ with ${\theta }_{22}=1/\left(w_{22}-{\text{w}}_{12}^{\text{T}}\beta \right)$.