A Monte Carlo-Based Fast Sample Entropy Algorithm
Zhengyang Huang, Chao Chen, Shizhen Huang · Frontiers in artificial intelligence and applications · 2024
Sample entropy is widely used in complexity analysis. It analyzes the complexity of time series by calculating the probability of template matching success at different scales. However, the direct calculation of sample entropy takes a long time. Although some studies have proposed several fast sample entropy algorithms, these algorithms still fail to achieve substantial acceleration with increasing data length. This paper proposes a fast sample entropy algorithm based on Monte Carlo (MC-LW algorithm). It is optimized based on a fast sample entropy algorithm—lightweight (LW algorithm) combined with the Monte Carlo method. First, the time series is downsampled using Monte Carlo methods, and then the optimized LW algorithm is used to calculate the downsampled time series. Direct sample entropy requires a similarity search of the full space. Fast algorithms also require a comparison of all-time series. This approach combines the advantages of the LW algorithm and Monte Carlo methods to accelerate the calculation of sample entropy. As the data length increases, the MC-LW algorithm can achieve more than a hundredfold acceleration compared to direct sample entropy algorithms. The paper experimentally compares the MC-LW algorithm, direct sample entropy algorithm, and LW algorithm in terms of entropy distinguishability and computation time, demonstrating that this algorithm effectively accelerates the calculation of sample entropy.