I have a dataframe containing customer data. Each row pertains to a different customer. One of the dataframe's series is 2016 sales to the customer. I'd like to categorizes the customers into deciles. Each decile will have one tenth of the total sales. Decile 10 will have the top customers which account for the first 10% of total sales.
What's the most python way to accomplish this task?
pandas has a pd.qcut method which, when applied to a Series, returns a categorical series (a DataFrame behaves similarly). So to get back a categorical index, you can do:
>>> import pandas as pd
# mock data
>>> df = pd.DataFrame({'sales_total': [1,2,162,126,126,12,7,1236,16,132,61,51]})
>>> cat_srs = pd.qcut(df['sales_total'], 10)
>>> print(cat_srs)
<<< 0 [1, 2.5]
1 [1, 2.5]
2 (159, 1236]
3 (1, 126]
4 (1, 126]
5 (8, 13.2]
6 (2.5, 8]
7 (159, 1236]
8 (13.2, 3]
9 (130.8, 159]
10 (56, 1]
11 (3, 56]
Name: dat, dtype: category
Categories (10, object): [...]
You can get the series entries with the values operator, and then from there get their bin code using the codes operator:
>>> deciles = cat_srs.values.codes
>>> print(deciles)
<<< array([0, 0, 9, 6, 6, 2, 1, 9, 3, 8, 5, 4], dtype=int8)
Which is what you need. From here you could assign the deciles to the data using df['decile'] = deciles, group entries using df.groupby('decile'), and so on.
The one liner for all of the above is pd.qcut(df['sales_total'], 10).values.codes.
Edit: answering the modified question below, per the comments—I don't know a way of doing this that's baked into a library. But assuming your data is relatively continuous, you can build classes yourself like so:
target = df['sales_total'].sum() / 10
deciles = []
sum = 0
classifier = 0
for val in df['sales_total']:
deciles.append(classifier)
sum += val
if sum > target:
classifier += 1
sum = 0
