I'm trying to find a good method for showing whether or not some variables are statistically significant explainers of a binary (0, 1) response in Python.
Catch is, I'd like to weight the data points by how important they are. Does anyone know of a good implementation in Python that allows for both weighted classification and statistical analysis of the fitted model?
Case weighted logistic regression gives some advice on working in SAS/R, but not python.
I don't think regression works here, as a good chunk of my data will just be zeroes. But I'd like to adjust the loss function to account for how impactful the data points are. If I'm missing something obvious let me know!
I've considered using a non-linear method like LightGBM that accepts a weights column but then I can't really do statistical tests - I'd just have to use feature importance scores. Far less scientific!
Thanks in advance!