为提升性能,应减少 pandas.DataFrame.apply() 的逐行操作,在本例中可改用 numpy.where() 二元操作符,如下In [1]: import pandas as pdIn [2]: import numpy as npIn [3]: df1 = pd.DataFrame({'t': [1,2,3], 'user_id': [10,20,30], 'v': [1.1,2.2,3.3]})In [4]: df1Out[4]:
t user_id v0 1 10 1.11 2 20 2.22 3 30 3.3In [5]: df2 = pd.DataFrame({'t': [4,1,2], 'user_id': [40,10,20], 'v': [400,100,200]})In [6]: df2Out[6]:
t user_id v0 4 40 4001 1 10 1002 2 20 200In [7]: df3 = pd.merge(df1, df2, how='right', on=['t', 'user_id'])In [8]: df3Out[8]:
t user_id v_x v_y0 1 10 1.1 1001 2 20 2.2 2002 4 40 NaN 400In [9]: df3['v'] = np.where(np.isnan(df3.v_x), df3.v_y, df3.v_x)In [10]: df3Out[10]:
t user_id v_x v_y v0 1 10 1.1 100 1.11 2 20 2.2 200 2.22 4 40 NaN 400 400.0In [11]: del df3['v_x']In [12]: del df3['v_y']In [13]: df3Out[13]:
t user_id v0 1 10 1.11 2 20 2.22 4 40 400.0