1. import pandas as pd
    2. import numpy as np
    5. def f1(a1, a2):
    6.     n = 0
    7.     for i in a1:
    8.         if i in a2:
    9.             n += 1
    10.     for i in a2:
    11.         if i in a1:
    12.             n += 1
    13.     return n
    16. def f3(a1, a2):
    17.     n = 0
    18.     n = map(lambda x, y: f1(x, y), a1, a2) 
    19.     return list(n)
    22. N = 5
    23. data = pd.DataFrame(np.random.uniform(1,9,(N,2)), columns=['A', 'B'], dtype=str)
    24. %timeit data1 = data['A'].combine(data['B'], f1)
    25. %timeit data1 = data.apply(lambda x: f1(x['A'], x['B']), axis=1)
    26. %timeit data1 = f3(data['A'], data['B'])
    27. %timeit data1 = [f1(x, y) for x, y in zip(data['A'], data['B'])]
    28. %timeit data1 = f3(data['A'].values, data['B'].values)
    29. # 218 µs ± 4.19 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
    30. # 698 µs ± 1.07 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
    31. # 37 µs ± 20.5 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)
    32. # 36.5 µs ± 26.5 ns per loop (mean ± std. dev. of 7 runs, 10000 loops each)
    33. # 16.9 µs ± 15.3 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)
    35. N = 500
    36. data = pd.DataFrame(np.random.uniform(1,9,(N,2)), columns=['A', 'B'], dtype=str)
    37. %timeit data1 = data['A'].combine(data['B'], f1)
    38. %timeit data1 = data.apply(lambda x: f1(x['A'], x['B']), axis=1)
    39. %timeit data1 = f3(data['A'], data['B'])
    40. %timeit data1 = [f1(x, y) for x, y in zip(data['A'], data['B'])]
    41. %timeit data1 = f3(data['A'].values, data['B'].values)
    42. # 13 ms ± 11.7 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
    43. # 11.2 ms ± 18.8 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
    44. # 1.02 ms ± 1.47 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
    45. # 982 µs ± 733 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
    46. # 1.01 ms ± 1 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
    48. N = 50000
    49. data = pd.DataFrame(np.random.uniform(1,9,(N,2)), columns=['A', 'B'], dtype=str)
    50. %timeit data1 = data['A'].combine(data['B'], f1)
    51. %timeit data1 = data.apply(lambda x: f1(x['A'], x['B']), axis=1)
    52. %timeit data1 = f3(data['A'], data['B'])
    53. %timeit data1 = [f1(x, y) for x, y in zip(data['A'], data['B'])]
    54. %timeit data1 = f3(data['A'].values, data['B'].values)
    55. # 1.29 s ± 3.74 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
    56. # 1.04 s ± 2.83 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
    57. # 99.6 ms ± 58.6 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
    58. # 96.7 ms ± 121 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

    显然，.apply方法效率并不高。