A classifier's true accuracy is p. You measure it on n independent examples. Return the probability the measured accuracy comes out at least threshold.

The number correct is $\mathrm{Binomial}(n, p)$, so sum from $\lceil n \cdot \text{threshold} \rceil$ upwards. On a small test set this is uncomfortably far from certain even when the model is genuinely good.

Input

n = 100
p = 0.9
threshold = 0.85

Output

0.9601094728889168