Question

I have a dataframe with numerical columns. For each column I would like calculate quantile information and assign each row to one of them. I tried to use the qcut() method to return a list of bins but instead ended up calculating the bins individually. What I thought might exist but I couldn’t find it would be a method like df.to_quintile(num of quantiles). This is what I came up with but I am wondering if there is a more succint/pandas way of doing this.

import pandas as pd

#create a dataframe
df = pd.DataFrame(randn(10, 4), columns=['A', 'B', 'C', 'D'])

def quintile(df, column):
    """
    calculate quintiles and assign each sample/column to a quintile 
    """
    #calculate the quintiles using pandas .quantile() here
    quintiles = [df[column].quantile(value) for value in [0.0,0.2,0.4,0.6,0.8]]
    quintiles.reverse() #reversing makes the next loop simpler

    #function to check membership in quintile to be used with pandas apply
    def check_quintile(x, quintiles=quintiles):
        for num,level in enumerate(quintiles):
            #print number, level, level[1]
            if  x >= level:
                print x, num
                return num+1

    df[column] = df[column].apply(check_quintile)

quintile(df,'A')

thanks,
zach cp

EDIT: After seeing DSMs answer the function can be written much simpler (below). Man, thats sweet.

def quantile(column, quantile=5):
    q = qcut(column, quantile)
    return len(q.levels)- q.labels
df.apply(quantile)
#or
df['A'].apply(quantile)

Answer 1

I think using the labels stored inside the Categorical object returned by qcut can make this a lot simpler. For example:

>>> import pandas as pd
>>> import numpy as np
>>> np.random.seed(1001)
>>> df = pd.DataFrame(np.random.randn(10, 2), columns=['A', 'B'])
>>> df
          A         B
0 -1.086446 -0.896065
1 -0.306299 -1.339934
2 -1.206586 -0.641727
3  1.307946  1.845460
4  0.829115 -0.023299
5 -0.208564 -0.916620
6 -1.074743 -0.086143
7  1.175839 -1.635092
8  1.228194  1.076386
9  0.394773 -0.387701
>>> q = pd.qcut(df["A"], 5)
>>> q
Categorical: A
array([[-1.207, -1.0771], (-1.0771, -0.248], [-1.207, -1.0771],
       (1.186, 1.308], (0.569, 1.186], (-0.248, 0.569], (-1.0771, -0.248],
       (0.569, 1.186], (1.186, 1.308], (-0.248, 0.569]], dtype=object)
Levels (5): Index([[-1.207, -1.0771], (-1.0771, -0.248],
                   (-0.248, 0.569], (0.569, 1.186], (1.186, 1.308]], dtype=object)
>>> q.labels
array([0, 1, 0, 4, 3, 2, 1, 3, 4, 2])

or to match your code:

>>> len(q.levels) - q.labels
array([5, 4, 5, 1, 2, 3, 4, 2, 1, 3])
>>> quintile(df, "A")
>>> np.array(df["A"])
array([5, 4, 5, 1, 2, 3, 4, 2, 1, 3])