2. import pandas as pd
    3. import numpy as np
    5. N = 10
    6. data = pd.DataFrame(np.random.uniform(1,9,(N,2)), columns=['A', 'B'])
    7. data1 = data.copy()
    8. data2 = data.copy()
    9. %time data1 = data.query('A > 4')
    10. %time data2 = data[data['A'] > 4]
    11. # CPU times: user 9.29 ms, sys: 3.84 ms, total: 13.1 ms
    12. # Wall time: 11.7 ms
    13. # CPU times: user 681 µs, sys: 0 ns, total: 681 µs
    14. # Wall time: 637 µs
    16. N = 1000
    17. data = pd.DataFrame(np.random.uniform(1,9,(N,2)), columns=['A', 'B'])
    18. data1 = data.copy()
    19. data2 = data.copy()
    20. %time data1 = data.query('A > 4')
    21. %time data2 = data[data['A'] > 4]
    22. # CPU times: user 988 µs, sys: 2.53 ms, total: 3.52 ms
    23. # Wall time: 2.64 ms
    24. # CPU times: user 1.64 ms, sys: 0 ns, total: 1.64 ms
    25. # Wall time: 1.18 ms
    27. N = 100000
    28. data = pd.DataFrame(np.random.uniform(1,9,(N,2)), columns=['A', 'B'])
    29. data1 = data.copy()
    30. data2 = data.copy()
    31. %time data1 = data.query('A > 4')
    32. %time data2 = data[data['A'] > 4]
    33. # CPU times: user 5.78 ms, sys: 2.95 ms, total: 8.73 ms
    34. # Wall time: 8.24 ms
    35. # CPU times: user 3.11 ms, sys: 781 µs, total: 3.89 ms
    36. # Wall time: 3.45 ms
    38. N = 10000000
    39. data = pd.DataFrame(np.random.uniform(1,9,(N,2)), columns=['A', 'B'])
    40. data1 = data.copy()
    41. data2 = data.copy()
    42. %time data1 = data.query('A > 4')
    43. %time data2 = data[data['A'] > 4]
    44. # CPU times: user 448 ms, sys: 416 ms, total: 864 ms
    45. # Wall time: 568 ms
    46. # CPU times: user 218 ms, sys: 173 ms, total: 391 ms
    47. # Wall time: 390 ms

    显然，data[data[‘A’] > 4] 较于 data.query(‘A > 4’) 更有优势。