A code to test different versions of shannon entropy calculation that were generated using Google Bard and Chat GPT. This is part of the work on [pyEntropy](https://github.com/nikdon/pyEntropy) library

benchmark_shannon.py

from collections import Counter

import numpy as np
import timeit


def shannon_entropy_original(time_series):
    """
    Return the Shannon Entropy of the sample data.

    Args:
        time_series: Vector or string of the sample data

    Returns:
        The Shannon Entropy as float value
    """

    # Check if string
    if not isinstance(time_series, str):
        time_series = list(time_series)

    # Create a frequency data
    data_set = list(set(time_series))
    freq_list = []
    for entry in data_set:
        counter = 0.0
        for i in time_series:
            if i == entry:
                counter += 1
        freq_list.append(float(counter) / len(time_series))

    # Shannon entropy
    ent = 0.0
    for freq in freq_list:
        ent += freq * np.log2(freq)
    ent = -ent
    return ent


def shannon_entropy_efficient(time_series):
    """
    Return the Shannon Entropy of the sample data.

    Args:
        time_series: Vector or string of the sample data

    Returns:
        The Shannon Entropy as float value
    """

    # BARD
    
    # # Check if string
    # if not isinstance(time_series, str):
    #     time_series = list(time_series)
    #
    # # Create a frequency data
    # data_set = set(time_series)
    # freq_list = np.array([float(time_series.count(entry)) / len(time_series) for entry in data_set])
    #
    # # Shannon entropy
    # ent = -np.sum(freq_list * np.log2(freq_list))
    # return ent

    # Chat GPT v1

    # # Check if string
    # if not isinstance(time_series, str):
    #     time_series = list(time_series)
    #
    # # Calculate frequency counts
    # counter = Counter(time_series)
    # total_count = len(time_series)
    #
    # # Calculate frequencies and Shannon entropy
    # ent = 0.0
    # for count in counter.values():
    #     freq = count / total_count
    #     ent += freq * np.log2(freq)
    #
    # ent = -ent
    # return ent
    
    # Chat GPT v2
    # Check if string
    if not isinstance(time_series, str):
        time_series = np.array(time_series)

    # Calculate frequency counts
    unique_vals, counts = np.unique(time_series, return_counts=True)
    total_count = len(time_series)

    # Calculate frequencies and Shannon entropy
    frequencies = counts / total_count
    ent = -np.sum(frequencies * np.log2(frequencies))

    return ent


if __name__ == "__main__":
    np.random.seed(42)

    # Define a list of orders and delays to test
    sizes = [10, 100, 500, 1000]

    summary = []

    for size in sizes:
        time_series = np.random.rand(size)

        Y_original = shannon_entropy_original(time_series)
        Y_efficient = shannon_entropy_efficient(time_series)

        if np.allclose(Y_original, Y_efficient, atol=1e-5):
            print(f"The outputs of the two versions are equal: {size=}")
        else:
            print(f"The outputs of the two versions are not equal: {size=}")

        number = 100
        repeat = 5

        stmt = f"shannon_entropy_original(time_series)"
        original_times = timeit.repeat(stmt, globals=globals(), number=number, repeat=repeat)
        original_avg_time = np.mean(original_times)

        stmt = f"shannon_entropy_efficient(time_series)"
        efficient_times = timeit.repeat(stmt, globals=globals(), number=number, repeat=repeat)
        efficient_avg_time = np.mean(efficient_times)

        time_difference = original_avg_time - efficient_avg_time
        percentage_difference = (time_difference / original_avg_time) * 100

        summary.append((size, original_avg_time, efficient_avg_time, percentage_difference))

    # Print the summary
    print("Size\tOriginal Time\tEfficient Time\tPercentage Difference")
    for entry in summary:
        size, original_time, efficient_time, percentage_difference = entry
        print(f"{size}\t{original_time:.6f}\t{efficient_time:.6f}\t{percentage_difference:.2f}%")