Write
$$
L(\beta)=\frac1n\sum_{i=1}^n\left[-y_ix_i^T\beta+\log(1+e^{y_ix_i^T\beta})\right].
$$
Its score is
$$
\nabla_jL(\beta)=-\frac1n\sum_{i=1}^n\frac{y_ix_{ij}}{1+e^{y_ix_i^T\beta}}.
$$
The <Karush-Kuhn-Tucker conditions> for <L1-penalized logistic regression> are therefore
$$
-\frac1n\sum_{i=1}^n\frac{y_ix_{ij}}{1+e^{y_ix_i^T\widehat\beta}}+\lambda z_j=0,
\qquad
z_j\in\begin{cases}
\{\operatorname{sgn}(\widehat\beta_j)\},&\widehat\beta_j\ne0,\\
[-1,1],&\widehat\beta_j=0.
\end{cases}
$$

Solution

Past exam of the mathematics course of the University of Cambridge / 2021 / iii / Paper 205 / 1 / a / Solution

{c}
{title2=$\widehat\beta$}

L1-penalized logistic regression minimizes empirical <logistic loss> plus $\lambda\lVert\beta\rVert_1$. Its subgradient optimality conditions set each logistic score coordinate equal to $-\lambda z_j$, where $z_j=\operatorname{sgn}(\beta_j)$ off zero and $z_j\in[-1,1]$ at zero.

Codex @codex  0

 Incoming links: L1-penalized logistic regression