Ex. 10.7

Ex. 10.7

Derive expression (10.32).

Soln. 10.7

We are looking for the optimum of

\[\begin{eqnarray} \hat \gamma_{jm} &=& \underset{\gamma_{jm}}{\operatorname{argmin}}\sum_{x_i\in R_{jm}}e^{-y_if_{m-1}(x_i)-y_i\gamma_{jm}}\non\\ &=&\underset{\gamma_{jm}}{\operatorname{argmin}}\sum_{x_i\in R_{jm}}w_i^{(m)}e^{-y_i\gamma_{jm}},\non \end{eqnarray}\]

where \(w_i^{(m)} = e^{-y_if_{m-1}(x_i)}\).

Let

\[\begin{equation} F(\gamma_{jm}) = \sum_{x_i\in R_{jm}}w_i^{(m)}e^{-y_i\gamma_{jm}}.\non \end{equation}\]

We have

\[\begin{eqnarray} \frac{\partial F}{\partial \gamma_{jm}} &=& \sum_{x_i\in R_{jm}}w_i^{(m)}e^{-y_i\gamma_{jm}}\cdot (-y_i)\non\\ &=&-\sum_{x_i\in R_{jm}, y_i = 1}w_i^{(m)}e^{-\gamma_{jm}} + \sum_{x_i\in R_{jm}, y_i = -1}w_i^{(m)}e^{\gamma_{jm}}.\non \end{eqnarray}\]

Setting it to be zero and solve for \(\gamma_{jm}\) we obtain

\[\begin{equation} \hat\gamma_{jm} = \frac{1}{2}\log \frac{\sum_{x_i\in R_{jm}}w_i^{(m)}\bb{1}(y_i=1)}{\sum_{x_i\in R_{jm}}w_i^{(m)}\bb{1}(y_i=-1)}.\non \end{equation}\]