#logistic-loss

A boosting algorithm gives example $t$ the unnormalized weight $-L'(z_t)$, where $z_t$ is the example's current margin $y_t h(x_t)$ and $L$ the margin loss. With the logistic loss $L(z) = \ln(1 + e^{-z})$, how many times larger is the weight of a badly misclassified example ($z = -5$) than that of a barely misclassified one ($z = -0.1$)?