[max.power]

The example with 1,000 data points and a tolerance of 0.05 on page 60 has a probability of >= 98%.

If I use the Hoeffding inequality with M = 1 and epsilon = 0.05, I get 98.7%.

If I use the generalization bound with delta = 0.05, I get 95.7%.

Is the example wrong or am I doing something horribly wrong?!