Capped / Constrained Weights in Python Pandas

I have a dataframe of weights, in which I want to constrain the maximum weight for any one element to 30%. However in doing this, the sum of the weights becomes less than 1... so the weights of all other elements should be uniformly increased, and then repetitively capped at 30% until the sum of all weights is 1.

for example:

If my data is in a pandas data frame, how can I do this efficiently?
Note: in reality I have like 20 elements which I want to cap at 10%... so there is much more processing involved. I also intent to run this step 1000s of times.

Thank you,
CWSE

2 Answers
2

Here's one vectorised solution. The idea is to calculate an adjustment and distribute it proportionately among the non-capped values.

df = pd.DataFrame('Elements': list('ABCDE'),
'Uncon': [0.53, 0.34, 0.06, 0.03, 0.03])

df['Con'] = np.minimum(0.30, df['Uncon'])
nonmax = df['Con'].ne(0.30)

adj = (1 - df['Con'].sum()) * df['Uncon'].loc[nonmax] / df['Uncon'].loc[nonmax].sum()

df['Con'] = df['Con'].mask(nonmax, df['Uncon'] + adj)

print(df)

Elements Uncon Con
0 A 0.53 0.3
1 B 0.34 0.3
2 C 0.06 0.2
3 D 0.03 0.1
4 E 0.03 0.1


@jpp

The following is a rough approach, modified from your answer to iteratively solveand re-cap. It doenst produce a perfect answer though... and having a while loop makes it inefficient. Any ideas how this could be improved?

import pandas as pd
import numpy as np
cap = 0.1
df = pd.DataFrame('Elements': list('ABCDEFGHIJKLMNO'),
'Values': [17,11,7,5,4,4,3,2,1.5,1,1,1,0.8,0.6,0.5])
df['Uncon'] = df['Values']/df['Values'].sum()
df['Con'] = np.minimum(cap, df['Uncon'])

while df['Con'].sum() < 1 or len(df['Con'][df['Con']>cap]) >=1:
df['Con'] = np.minimum(cap, df['Con'])
nonmax = df['Con'].ne(cap)

adj = (1 - df['Con'].sum()) * df['Con'].loc[nonmax] /
df['Uncon'].loc[nonmax].sum()

df['Con'] = df['Con'].mask(nonmax, df['Con'] + adj)
print(df)
print(df['Con'].sum())


By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.