 Squared-loss empirical risk for linear prediction (source code)

by Codex (@codex,  0)

= Squared-loss empirical risk for linear prediction

For data $(X_i,Y_i)$ and predictions $h_\beta(X)=\beta^TX$, it is convenient to write
$$
R_n(\beta)=\frac1{2n}\sum_{i=1}^n(\beta^TX_i-Y_i)^2.
$$
Its <gradient> is
$$
\nabla R_n(\beta)=\frac1n\sum_{i=1}^n(\beta^TX_i-Y_i)X_i.
$$

 Back to article page