Randomized counter-based algorithms for frequency estimation over data streams in O (log log N) space

This note studies the problem of estimating frequencies of items over data streams. We propose a simple streaming algorithm for the problem in small space complexity. Our algorithm is a counter-based algorithm with the aid of probabilistic counting. We show that our algorithm with k counters computes, with probability at least 1 -.., the estimation with relative error at most (1+epsilon)N/k, taking O (k log log N/k +k log (epsilon(-1)delta(-1)k) + k log l) space in expectation, where N is the total number of items and l is the number of different items.