Multiple Sample Weight in BF

[henrydingliu]

Currently in test_workflow, there's a broadcast warning because the test would try to fit BF on both incurred and paid. this can be replicated thus.

clrd = cl.load_sample('clrd')
aegis = clrd.loc['Aegis Grp']
cl.BornhuetterFerguson().fit(aegis[['IncurLoss','CumPaidLoss']],sample_weight=aegis['EarnedPremCeded'].latest_diagonal)

/local_disk0/.ephemeral_nfs/envs/pythonEnv-a0afa165-3cc7-4e93-a3bd-112ea002bc32/lib/python3.12/site-packages/chainladder/methods/base.py:160: UserWarning: X and sample_weight are not aligned. Broadcasting may occur.

it would be simple enough to change one of the slicing operators to broadcast explicitly. but then I got thinking, it's entirely realistic for users to want to have different sample_weight, for example, if they want to get ultimate loss and ultimate count in a single BF estimator, and want to use premium as the sample weight for loss and exposure as the sample weight for count. however, this is currently not supported.

cl.BornhuetterFerguson().fit(aegis[['IncurLoss','CumPaidLoss']],sample_weight=aegis[['EarnedPremDIR','EarnedPremCeded']].latest_diagonal)

ValueError: operands could not be broadcast together with shapes (3,4,10,1) (3,2,10,1) 

Looks like BF is trying to join between columns from the triangle to columns from sample weight.

Do people actually want to have different sample_weights inside BF (and benktander by extension)?