qcut中bin的对称数目大约为零

时间:2018-07-19 21:49:40

标签: python pandas binning

我有一个pandas数据帧,该数据帧具有不同数量的整数,并且每行中有NaNs。我想将每行中的值分配为8个bin-每行4个bin负值和4个正面值。因此,每行每个bin中将有不同数量的值。关于如何调整qcut功能的任何提示?谢谢!

1 个答案:

答案 0 :(得分:0)

如果我理解正确,您可以对正值进行qcut,对负值进行qcut

例如,给定数据框:

>>> df
        vals
0  -0.456460
1   0.448368
2   0.186750
3   1.056617
4  -0.035620
5  -0.609843
6   0.126376
7   0.160817
8  -1.495441
9   0.730763
10 -0.005071
11  0.677918
12 -0.779553
13  0.717374
14  2.250258
15 -0.801028
16  0.306408
17  0.538970
18 -2.120528
19  1.066903

使用2 qcuts,一个为正,另一个为负。

df.loc[df.vals > 0,'bin'] = pd.qcut(df.loc[df.vals > 0,'vals'], q=4)

df.loc[df.vals < 0,'bin'] = pd.qcut(df.loc[df.vals < 0,'vals'], q=4)

因此,它们被分为8个唯一的bin,其中4个为正,4个为负:

>>> df
        vals                 bin
0  -0.456460    (-0.695, -0.351]
1   0.448368      (0.276, 0.608]
2   0.186750      (0.125, 0.276]
3   1.056617       (0.812, 2.25]
4  -0.035620  (-0.351, -0.00507]
5  -0.609843    (-0.695, -0.351]
6   0.126376      (0.125, 0.276]
7   0.160817      (0.125, 0.276]
8  -1.495441    (-2.122, -0.975]
9   0.730763      (0.608, 0.812]
10 -0.005071  (-0.351, -0.00507]
11  0.677918      (0.608, 0.812]
12 -0.779553    (-0.975, -0.695]
13  0.717374      (0.608, 0.812]
14  2.250258       (0.812, 2.25]
15 -0.801028    (-0.975, -0.695]
16  0.306408      (0.276, 0.608]
17  0.538970      (0.276, 0.608]
18 -2.120528    (-2.122, -0.975]
19  1.066903       (0.812, 2.25]

您可以对垃圾箱进行排序以使其可视化,这样您就可以看到4个正值箱和4个负值箱:

np.sort(df['bin'].unique())

array([Interval(-2.1219999999999999, -0.97499999999999998, closed='right'),
       Interval(-0.97499999999999998, -0.69499999999999995, closed='right'),
       Interval(-0.69499999999999995, -0.35099999999999998, closed='right'),
       Interval(-0.35099999999999998, -0.0050699999999999999, closed='right'),
       Interval(0.125, 0.27600000000000002, closed='right'),
       Interval(0.27600000000000002, 0.60799999999999998, closed='right'),
       Interval(0.60799999999999998, 0.81200000000000006, closed='right'),
       Interval(0.81200000000000006, 2.25, closed='right')], dtype=object)