FLAME JMLR 2020 Experiment Code

This is the repository for the paper FLAME: A Fast Large-scale Almost Matching Exactly Approach to Causal Inference (JMLR 2020). This contains the scripts for the experiment section in the paper. Detailed instructions are also in the comments of the scripts.

Please note that this repository is not intended for users. There are user-facing R and Python packages available in the repositories https://github.com/almost-matching-exactly/R-FLAME and https://github.com/almost-matching-exactly/DAME-FLAME-Python-Package, respectively.

FLAMEbit example

from FLAMEbit import *

df,_,_ = data_generation_dense_2(15000, 15000, 10, 5) # data generation
holdout,_,_ = data_generation_dense_2(15000, 15000, 10, 5) # data generation (the holdout set)

res = run_bit(df = df, holdout = holdout, covs = range(15), covs_max_list = [2]*15, tradeoff_param = 0.1) % call the function

estimate, group_size = get_estimate_vectors(df, res[1], range(15)) # get result summary

Note

The columns of the data table ''df'' must be reordered such that the =covs_max_list= is non-increasing from left to right.