Write a function `auto_avg` that returns the differentially private average of a Pandas series `s`. Your function should automatically determine the clipping parameter `b`, and should enforce differential privacy for a **total privacy cost** of `epsilon`. You can assume that all values are non-negative (i.e. 0 or greater).

# IMPLEMENT YOUR CODE HERE raise NotImplementedError()

# TEST CASE r1 = np.mean([pct_error(adult['Age'].mean(), auto_avg(adult['Age'], 1.0)) for _ in range(20)]) r2 = np.mean([pct_error(adult['Capital Gain'].mean(), auto_avg(adult['Capital Gain'], 1.0)) for _ in range(20)]) r3 = np.mean([pct_error(adult['fnlwgt'].mean(), auto_avg(adult['fnlwgt'], 1.0)) for _ in range(20)]) print('Average errors:', r1, r2, r3) assert r1 > 0 assert r2 > 0 assert r3 > 0 assert r1 < 1 assert r2 < 100 assert r3 < 1