Sort bins from pandas cut

问题:

Using pandas cut I can define bins by providing the edges and pandas creates bins like (a, b].

My question is how can I sort the bins (from the lowest to the highest)?

import numpy as np
import pandas as pd

y = pd.Series(np.random.randn(100))

x1 = pd.Series(np.sign(np.random.randn(100)))
x2 = pd.cut(pd.Series(np.random.randn(100)), bins = [-3, -0.5, 0, 0.5, 3])

model = pd.concat([y, x1, x2], axis = 1, keys = ['Y', 'X1', 'X2'])

I have an intermediate result where the order of the bins is preserved

int_output = model.groupby(['X1', 'X2']).mean().unstack()
int_output.columns = int_output.columns.get_level_values(1)

X2    (-3, -0.5]  (-0.5, 0]  (0, 0.5]  (0.5, 3]
X1                                             
-1.0    0.101475  -0.344419 -0.482992 -0.015179
 1.0    0.249961   0.484757 -0.066383 -0.249414

But then I do other operations that arbitrarily changes the order of the bins:

output = pd.concat(int_output.to_dict('series'), axis = 1)

      (-0.5, 0]  (-3, -0.5]  (0, 0.5]  (0.5, 3]
X1                                             
-1.0  -0.344419    0.101475 -0.482992 -0.015179
 1.0   0.484757    0.249961 -0.066383 -0.249414

Now I would like to plot the data in a bar chart, but I want the bins to be sorted from the lowest (-3, -0.5] to the highest (0.5, 3].

I think I can achieve this by manipulating the string, using a split on "," and then cleaning brackets, but I would like to know if there is a better way.

回答1:

There is main problem losing ordered CategoricalIndex.

np.random.seed(12456)
y = pd.Series(np.random.randn(100))
x1 = pd.Series(np.sign(np.random.randn(100)))
x2 = pd.cut(pd.Series(np.random.randn(100)), bins = [-3, -0.5, 0, 0.5, 3])

model = pd.concat([y, x1, x2], axis = 1, keys = ['Y', 'X1', 'X2'])
int_output = model.groupby(['X1', 'X2']).mean().unstack()
int_output.columns = int_output.columns.get_level_values(1)

print (int_output)
X2    (-3, -0.5]  (-0.5, 0]  (0, 0.5]  (0.5, 3]
X1                                             
-1.0    0.230060  -0.079266 -0.079834 -0.064455
 1.0   -0.451351   0.268688  0.020091 -0.280218

print (int_output.columns)
CategoricalIndex(['(-3, -0.5]', '(-0.5, 0]', '(0, 0.5]', '(0.5, 3]'], 
                 categories=['(-3, -0.5]', '(-0.5, 0]', '(0, 0.5]', '(0.5, 3]'], 
                 ordered=True, name='X2', dtype='category')

output = pd.concat(int_output.to_dict('series'), axis = 1)
print (output)
      (-0.5, 0]  (-3, -0.5]  (0, 0.5]  (0.5, 3]
X1                                             
-1.0  -0.079266    0.230060 -0.079834 -0.064455
 1.0   0.268688   -0.451351  0.020091 -0.280218

print (output.columns)
Index(['(-0.5, 0]', '(-3, -0.5]', '(0, 0.5]', '(0.5, 3]'], dtype='object')

One possible solution is extract first number from output.columns, create helper Series and sort it. Last reindex original columns:

cat = output.columns.str.extract('\((.*),', expand=False).astype(float)
a = pd.Series(cat, index=output.columns).sort_values()
print (a)
(-3, -0.5]   -3.0
(-0.5, 0]    -0.5
(0, 0.5]      0.0
(0.5, 3]      0.5
dtype: float64

output = output.reindex(columns=a.index)
print (output)
      (-3, -0.5]  (-0.5, 0]  (0, 0.5]  (0.5, 3]
X1                                             
-1.0    0.230060  -0.079266 -0.079834 -0.064455
 1.0   -0.451351   0.268688  0.020091 -0.280218

回答2:

An easy fix to the problem you've highlighted above is to simply reorder the columns:

output[sorted(output.columns)]

回答3:

I made a function to do so.

def dfsortbybins(df, col):
    """
    param df: pandas dataframe
    param col: name of column containing bins
    """
    d=dict(zip(bins,[float(s.split(',')[0].split('(')[1]) for s in bins]))
    df[f'{col} dfrankbybins']=df.apply(lambda x : d[x[col]] if not pd.isnull(x[col]) else x[col], axis=1)
    df=df.sort_values(f'{col} dfrankbybins').drop(f'{col} dfrankbybins',axis=1)
    return df